Home

Iceberg: A loudspeaker-based room

auralization method for auditory

research

Sergio Luiz Aguirre

Submitted in fulﬁlment of the requirements for the degree of

Doctor of Philosophy

Hearing Sciences – Scottish Section - School of Medicine

University of Nottingham

Supervised by William M. Whitmer, Lars Bramsløw, &

Graham Naylor

2022

There is no “nonspatial hearing” — Jens Blauert

Abstract

Depending on the acoustic scenario, people with hearing loss are challenged

on a diﬀerent scale than normal hearing people to comprehend sound, espe-

cially speech. That happen especially during social interactions within a group,

which often occurs in environments with low signal-to-noise ratios. This com-

munication disruption can create a barrier for people to acquire and develop

communication skills as a child or to interact with society as an adult. Hear-

ing loss compensation aims to provide an opportunity to restore the auditory

part of socialization. Technology and academic eﬀorts progressed to a bet-

ter understanding of the human hearing system. Through constant eﬀorts

to present new algorithms, miniaturization, and new materials, constantly-

improving hardware with high-end software is being developed with new fea-

tures and solutions to broad and speciﬁc auditory challenges. The eﬀort to

deliver innovative solutions to the complex phenomena of hearing loss encom-

passes tests, veriﬁcations, and validation in various forms. As the newer de-

vices achieve their purpose, the tests need to increase the sensitivity, requiring

conditions that eﬀectively assess their improvements.

Regarding realism, many levels are required in hearing research, from pure

tone assessment in small soundproof booths to hundreds of loudspeakers com-

bined with visual stimuli through projectors or head-mounted displays, light,

and movement control. Hearing aids research commonly relies on loudspeaker

setups to reproduce sound sources. In addition, auditory research can use

well-known auralization techniques to generate sound signals. These signals

can be encoded to carry more than sound pressure level information, adding

spatial information about the environment where that sound event happened

or was simulated. This work reviews physical acoustics, virtualization, and

auralization concepts and their uses in listening eﬀort research. This knowl-

edge, combined with the experiments executed during the studies, aimed to

provide a hybrid auralization method to be virtualized in four-loudspeaker se-

tups. Auralization methods are techniques used to encode spatial information

into sounds. The main methods were discussed and derived, observing their

spatial sound characteristics and trade-oﬀs to be used in auditory tests with

one or two participants. Two well-known auralization techniques (Ambisonics

and Vector-Based Amplitude Panning) were selected and compared through

a calibrated virtualization setup regarding spatial distortions in the binau-

ral cues. The choice of techniques was based on the need for loudspeakers,

although a small number of them. Furthermore, the spatial cues were exam-

ined by adding a second listener to the virtualized sound ﬁeld. The outcome

reinforced the literature around spatial localization and these techniques driv-

ing Ambisonics to be less spatially accurate but with greater immersion than

Vector-Based Amplitude Panning.

A combination study to observe changes in listening eﬀort due to diﬀerent

signal-to-noise ratios and reverberation in a virtualized setup was deﬁned. This

experiment aimed to produce the correct sound ﬁeld via a virtualized setup

and assess listening eﬀort via subjective impression with a questionnaire, an

objective physiological outcome from EEG, and behavioral performance on

word recognition. Nine levels of degradation were imposed on speech signals

over speech maskers separated in the virtualized space through Ambisonics’

ﬁrst-order technique in a setup with 24 loudspeakers. A high correlation be-

tween participants’ performance and their responses on the questionnaire was

observed. The results showed that the increased virtualized reverberation time

negatively impacts speech intelligibility and listening eﬀort.

A new hybrid auralization method was proposed merging the investigated tech-

niques that presented complementary spatial sound features. The method was

derived through room acoustics concepts and a speciﬁc objective parameter

derived from the room impulse response called Center Time. The veriﬁcation

around the binaural cues was driven with three diﬀerent rooms (simulated).

As the validation with test subjects was not possible due to the COVID-19

pandemic situation, a psychoacoustic model was implemented to estimate the

spatial accuracy of the method within a four-loudspeaker setup. Also, an in-

vestigation ran the same veriﬁcation, and the model estimation was performed

with the introduction of hearing aids. The results showed that it is possible

to consider the hybrid method with four loudspeakers for audiological tests

while considering some limitations. The setup can provide binaural cues to a

maximum ambiguity angle of 30 degrees in the horizontal plane for a centered

listener.

Acknowledgements

Thank you,

Obrigado, Gracias, Grazie, Tak skal du have, Dank u zeer & Danke sehr

Firstly, I would like to express my gratitude to my supervisors, Drs. Bill Whit-

mer, Lars Bramsløw, and Graham Naylor, for their guidance, expertise, and

remarkable patience throughout this process. Your support and mentorship

have been invaluable in helping me to ﬁll my knowledge gaps, tirelessly encour-

aging me to ask the right questions, and guiding me to produce high-quality

scientiﬁc research. I also thank Dr. Thomas Lunner for his initial guidance,

insightful questions, and comments.

Thank you to the special people at Eriksholm Research Centre in Denmark

and the people at Hearing Sciences – Scottish Section. Working with such

fantastic top teams has been a joy and a privilege. A special thank you to Jette,

Michael, Bo, Niels, Claus, Dorothea, Sergi, Jeppe, James, Lorenz, Johannes,

and Hamish. Thanks to all the people involved in HEAR-ECO for their hard

work, especially Hidde, Beth, Patrycja, Tirdad, and Defne.

I am deeply grateful to my sweet wife, Lilian, for the love, encouragement, and

support she has given me throughout this journey. Her unwavering support

has been an enduring seed of resilience and inspiration. I cannot thank her

enough for being such an integral part of my life.

Thanks to my friends Math and Gil, for always being there. Special thanks to

my former professors, Drs.: Arcanjo Lenzi, William D’Andrea Fonseca, Eric

Brand˜ao, Paulo Mareze, Stephan Paul, and Bruno Sanches Masiero for stimu-

lating critical thinking and for all the support, knowledge, and encouragement.

Thank you also to my former Oticon Medical colleagues Simon Ziska Krogholt,

Patrick Maas, Brian Skov, and Jens T. Balslev for the and support.

I would like to thank you, my Professor, Professora Dra. Dinara Xavier Paix˜ao.

All of this is possible because of you and your determination to create an

oﬃcial undergraduate course in Acoustical Engineering in Brazil. This course

iii

is praised for forming remarkable professionals who are recognized worldwide.

It is not just my dream that you have made possible, but the countless people

for whom this course has been a life-changing experience. We know that this

was a collective eﬀort, but your role was vital. Your way of showing that

politics is a part of everything, and that we need to be gentle but correct,

made all the diﬀerence. Thank you. Muito Obrigado.

I sincerely thank my friends and colleagues from my undergraduate studies

in acoustical engineering (USFM/EAC) and the master’s (UFSC/LVA). Your

support, happiness, patience and encouragement have been invaluable through-

out this journey. Thank you for helping me to develop my skills and knowledge

and for being such a positive inﬂuence on my academic career. I am deeply

grateful for all you have done for me and look forward to continuing our pro-

fessional relationship. I want to thank the oldest friends, Fabrício, André, Juliano,

and the Panteon. I value the bond and history that we share. Thank you for

being such wonderful friends.

I want to express my heartfelt thanks to the Brazilian CNPq and the govern-

ment (Lula/Dilma) policies that support students with low income and from

public schools. With their ﬁnancial assistance, I could pursue my studies and

achieve my goals. I am deeply grateful for their support and the opportunity to

receive a quality education. I would also like to express my gratitude to Marie

Sk lodowska-Curie Actions for their support of my doctoral education. Their

reference programme for doctoral education has provided me with invaluable

resources and opportunities, and I am extremely grateful for their support.

Thank you for helping me achieve my goals and being a valuable part of my

academic journey.

I want to express my gratitude to all those who will read this thesis in the

future. Your time and attention are greatly appreciated. I wish you a good

reading experience and hope that you will ﬁnd the ideas and research presented

in this work to be both thought-provoking and beneﬁcial. Thank you again

for considering this work.

Author’s Declaration

This thesis is the result of the author’s original research. Chapter 4 is a

collaboration work with Tirdad Seiﬁ-Ala. The author has composed it and

has not been previously submitted for any other academic qualiﬁcation.

This project has received funding from the European Union’s Horizon 2020

research and innovation programme under the Marie-Sk lodowska-Curie grant

agreement No 765329; The funder had no role in study design.

Sergio Luiz Aguirre

Contents

Abstract i

Acknowledgements iii

Author’s Declaration v

Nomenclature xxi

1 Introduction 1

1.1 Motivations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Aims and Scope . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.3 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.4 Organization of the Thesis . . . . . . . . . . . . . . . . . . . . . 4

2 Literature Review 7

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.2 Human Binaural Hearing . . . . . . . . . . . . . . . . . . . . . . 8

2.2.1 Spatial Hearing Concepts . . . . . . . . . . . . . . . . . 9

2.2.2 Binaural cues . . . . . . . . . . . . . . . . . . . . . . . . 10

2.2.3 Monaural cues . . . . . . . . . . . . . . . . . . . . . . . . 11

CONTENTS vii

2.2.4 Head-related transfer function . . . . . . . . . . . . . . . 12

2.2.5 Subjective aspects of an audible reﬂection . . . . . . . . 16

2.3 Spatial Sound & Virtual Acoustics . . . . . . . . . . . . . . . . 17

2.3.1 Virtualization . . . . . . . . . . . . . . . . . . . . . . . . 18

2.3.1.1 Auralization . . . . . . . . . . . . . . . . . . . . 19

2.3.1.2 Reproduction . . . . . . . . . . . . . . . . . . . 21

2.3.2 Auralization Paradigms . . . . . . . . . . . . . . . . . . 22

2.3.2.1 Binaural . . . . . . . . . . . . . . . . . . . . . . 22

2.3.2.2 Panorama . . . . . . . . . . . . . . . . . . . . . 23

2.3.2.3 Sound Field Synthesis . . . . . . . . . . . . . . 31

2.3.3 Room acoustics . . . . . . . . . . . . . . . . . . . . . . . 32

2.3.3.1 Room acoustics parameters . . . . . . . . . . . 33

2.3.3.2 Reverberation Time . . . . . . . . . . . . . . . 33

2.3.3.3 Clarity and Deﬁnition . . . . . . . . . . . . . . 35

2.3.3.4 Center Time . . . . . . . . . . . . . . . . . . . 37

2.3.3.5 Parameters related to spatiality . . . . . . . . . 37

2.3.4 Loudspeaker-based Virtualization in Auditory Research . 40

2.3.4.1 Hybrid Methods . . . . . . . . . . . . . . . . . 43

2.3.4.2 Sound Source Localization . . . . . . . . . . . . 45

2.4 Listening Eﬀort Assessment . . . . . . . . . . . . . . . . . . . . 50

2.5 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . 53

3 Binaural cue distortions in virtualized Ambisonics and VBAP 55

CONTENTS viii

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

3.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

3.2.1 Setups and system characterization . . . . . . . . . . . . 59

3.2.1.1 Reverberation time . . . . . . . . . . . . . . . . 60

3.2.1.2 Early-reﬂections . . . . . . . . . . . . . . . . . 61

3.2.2 Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . 62

3.2.3 Calibration . . . . . . . . . . . . . . . . . . . . . . . . . 64

3.2.4 VBAP Auralization . . . . . . . . . . . . . . . . . . . . . 67

3.2.5 Ambisonics Auralization . . . . . . . . . . . . . . . . . . 67

3.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

3.3.1 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

3.3.2 Centered position . . . . . . . . . . . . . . . . . . . . . . 71

3.3.2.1 Centered ITD . . . . . . . . . . . . . . . . . . . 73

3.3.2.2 Centered ILD . . . . . . . . . . . . . . . . . . . 76

3.3.3 Oﬀ-centered position . . . . . . . . . . . . . . . . . . . . 82

3.3.3.1 Oﬀ-center ITD . . . . . . . . . . . . . . . . . . 83

3.3.3.2 Oﬀ-center ILD . . . . . . . . . . . . . . . . . . 87

3.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

3.5 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . 95

4 Subjective Eﬀort within Virtualized Sound Scenarios 97

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98

4.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100

CONTENTS ix

4.2.1 Participants . . . . . . . . . . . . . . . . . . . . . . . . . 102

4.2.2 Stimuli . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102

4.2.3 Apparatus . . . . . . . . . . . . . . . . . . . . . . . . . . 104

4.2.4 Auralization . . . . . . . . . . . . . . . . . . . . . . . . . 107

4.2.5 Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . 110

4.2.6 Questionnaire . . . . . . . . . . . . . . . . . . . . . . . . 112

4.2.7 Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

4.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

4.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116

4.5 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . 119

5 Iceberg: A Hybrid Auralization Method Focused on Compact

Setups. 120

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120

5.2 Iceberg an Hybrid Auralization Method . . . . . . . . . . . . . . 122

5.2.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . 122

5.2.2 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . 123

5.2.2.1 Components . . . . . . . . . . . . . . . . . . . . 124

5.2.2.2 Energy Balance . . . . . . . . . . . . . . . . . . 124

5.2.2.3 Iceberg proposition . . . . . . . . . . . . . . . . 127

5.2.3 Setup Equalization & Calibration: . . . . . . . . . . . . . 133

5.3 System Characterization . . . . . . . . . . . . . . . . . . . . . . 138

5.3.1 Experimental Setup . . . . . . . . . . . . . . . . . . . . . 139

CONTENTS x

5.3.2 Virtualized RIRs & BRIRs . . . . . . . . . . . . . . . . . 139

5.3.3 Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . 142

5.3.4 Reverberation Time . . . . . . . . . . . . . . . . . . . . . 143

5.4 Main Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145

5.4.1 Centered Position . . . . . . . . . . . . . . . . . . . . . . 145

5.4.1.1 Interaural Time Diﬀerence . . . . . . . . . . . . 145

5.4.1.2 Interaural Level Diﬀerence . . . . . . . . . . . . 147

5.4.1.3 Azimuth Estimation . . . . . . . . . . . . . . . 150

5.4.2 Oﬀ-Center Positions . . . . . . . . . . . . . . . . . . . . 151

5.4.2.1 Interaural Time Diﬀerence . . . . . . . . . . . . 151

5.4.2.2 Interaural Level Diﬀerence . . . . . . . . . . . . 153

5.4.2.3 Azimuth Estimation . . . . . . . . . . . . . . . 155

5.4.3 Centered Accompanied by a Second Listener . . . . . . . 156

5.4.3.1 Interaural Time Diﬀerence . . . . . . . . . . . . 157

5.4.3.2 Interaural Level Diﬀerence . . . . . . . . . . . . 157

5.4.3.3 Azimuth Estimation . . . . . . . . . . . . . . . 159

5.5 Supplementary Test Results . . . . . . . . . . . . . . . . . . . . 163

5.5.1 Centered Position (Aided) . . . . . . . . . . . . . . . . . 164

5.5.1.1 Interaural Time Diﬀerence . . . . . . . . . . . . 165

5.5.1.2 Interaural Level Diﬀerence . . . . . . . . . . . . 166

5.5.1.3 Azimuth Estimation . . . . . . . . . . . . . . . 166

5.5.2 Oﬀ-center Positions (Aided) . . . . . . . . . . . . . . . . 169

CONTENTS xi

5.5.2.1 Interaural Time Diﬀerence . . . . . . . . . . . . 169

5.5.2.2 Interaural Level Diﬀerence . . . . . . . . . . . . 171

5.5.2.3 Azimuth Estimation . . . . . . . . . . . . . . . 174

5.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176

5.6.1 Subjective impressions . . . . . . . . . . . . . . . . . . . 180

5.6.2 Advantages and Limitations . . . . . . . . . . . . . . . . 181

5.6.3 Study limitations and Future Work . . . . . . . . . . . . 182

5.7 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . 183

6 Conclusion 184

6.1 Iceberg . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185

6.2 General Discussion . . . . . . . . . . . . . . . . . . . . . . . . . 187

6.2.1 Iceberg capabilities . . . . . . . . . . . . . . . . . . . . . 189

6.2.2 Iceberg & Second Joint Listener . . . . . . . . . . . . . . 190

6.2.3 Iceberg: Listener Wearing Hearing Aids . . . . . . . . . . 191

6.2.4 Iceberg Limitations . . . . . . . . . . . . . . . . . . . . . 192

6.3 General Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . 193

6.4 Main Contributions . . . . . . . . . . . . . . . . . . . . . . . . . 193

Bibliography 195

Appendices 225

A ITDs Ambisonics 225

B Delta ILD Ambisonics 226

C Wave Equation and Spherical Harmonic Representation 228

C.1 Wave Equation in Spherical Coordinates . . . . . . . . . . . . . 228

C.2 Separation of the Variables . . . . . . . . . . . . . . . . . . . . . 228

C.3 Spherical Harmonics . . . . . . . . . . . . . . . . . . . . . . . . 230

D Reverberation time in Acoustic Simulation 231

E Alpha Coeﬃcients 232

F Questionnaire 234

xii

List of Tables

2.1 Non-exhaustive overview list of hybrid auralization methods

proposed in the literature. The A-B order of the techniques

does not represent any order of signiﬁcance. . . . . . . . . . . . 45

2.2 Overview of Localization Error Estimates or Measurements from

Loudspeaker-Based Virtualization Systems Using Various Au-

ralization Methods. . . . . . . . . . . . . . . . . . . . . . . . . . 49

3.1 Sound pressure level diﬀerence between direct sound and early

reﬂections ∆ SPL [dB] . . . . . . . . . . . . . . . . . . . . . . . 62

4.1 The questionnaire for subjective ratings of performance, eﬀort

and engagement (English translation from Danish) . . . . . . . 113

4.2 Results of linear mixed model based on SNR and RT predictors

estimates of the questionnaire. . . . . . . . . . . . . . . . . . . . 114

4.3 Pearson skipped correlations between performance and self-reported

questions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116

5.1 Reverberation Time in three virtualized environments . . . . . . 144

5.2 One way anova, columns are absolute diﬀerence between esti-

mated and reference angles for diﬀerent KEMAR positions and

RTs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162

5.3 Hearing Level in dB according to the proposed Standard Au-

diograms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164

5.4 One way anova, columns are absolute diﬀerence between esti-

mated and reference angles for diﬀerent positions and RTs. . . . 168

xiii

5.5 Maximum ITD according to displacement . . . . . . . . . . . . 171

xiv

List of Figures

2.1 Two-dimensional representation of the cone of confusion. . . . . 11

2.2 A descriptive deﬁnition of the measured free-ﬁeld HRTF for a

given angle. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

2.3 Polar coordinate system related to head incidence angles . . . . 13

2.4 Head-related transfer functions of four human test participants,

frontal incidence . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

2.5 Audible eﬀects of a single reﬂection . . . . . . . . . . . . . . . . 16

2.6 Binaural reproduction setups . . . . . . . . . . . . . . . . . . . . 23

2.7 Vector-based amplitude panning: 2D display of sound sources

positions and weights. . . . . . . . . . . . . . . . . . . . . . . . 25

2.8 Diagram representing the placement of speakers in the VBAP

technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

2.9 Spherical Harmonics Y

(θ, ϕ). . . . . . . . . . . . . . . . . . . . 29

2.10 B-format components: omnidirectional pressure component W,

and the three velocity components X, Y, Z. . . . . . . . . . . . . 30

2.11 Illustration of Huygen’s Principle of a propagating wave front. . 32

2.12 Normalized Room Impulse Response: example from a real room

in the time domain (left), and in the time domain in dB (right). 35

2.13 LoRa implementation processing diagram . . . . . . . . . . . . . 43

3.1 Hearing Sciences - Scottish Section Test Room. . . . . . . . . . 59

LIST OF FIGURES xvi

3.2 Eriksholm Test Room. . . . . . . . . . . . . . . . . . . . . . . . 59

3.3 Reverberation time in third of octave . . . . . . . . . . . . . . . 61

3.4 HATS and KEMAR inside test room in Glasgow . . . . . . . . . 63

3.5 HATS and KEMAR inside Eriksholm’s Anechoic Room . . . . 63

3.6 Description of experiment’s measured positions and mannequin

placement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

3.7 Interaural cross correlation - Frontal angle . . . . . . . . . . . . 72

3.8 Polar representation IACC . . . . . . . . . . . . . . . . . . . . . 72

3.9 Interaural Time Diﬀerence by angle: VBAP accompanied . . . . 74

3.10 Ambisonics - directivity representation in 2D . . . . . . . . . . 75

3.11 Interaural Time Diﬀerence by angle: Ambisoncis accompanied . 76

3.12 Interaural Level Diﬀerences: VBAP and Ambisonics . . . . . . . 77

3.13 Interaural Level Diﬀerences: averaged octave bands as a func-

tion of azimuth angle for a HATS Br¨uel and Kjær TYPE 4128-C

in the horizontal plane. . . . . . . . . . . . . . . . . . . . . . . . 78

3.14 Interaural Level Diﬀerences with additional listener (VBAP and

Ambisonics) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

3.15 Discrepancies in Interaural Level Diﬀerences (VBAP) . . . . . . 79

3.16 VBAP Interaural Level Diﬀerences as function of azimuth angle

around the centered listener. . . . . . . . . . . . . . . . . . . . . 80

3.17 Discrepancies in Interaural Level Diﬀerences: Ambisonics . . . . 81

3.18 Ambisonics Interaural Level Diﬀerences as function of azimuth

angle around the centered listener. . . . . . . . . . . . . . . . . 82

3.19 VBAP Oﬀ center ITD HATS 25 cm . . . . . . . . . . . . . . . . 84

3.20 VBAP Oﬀ center ITD HATS 50 cm . . . . . . . . . . . . . . . . 84

3.21 VBAP Oﬀ center ITD HATS 75 cm . . . . . . . . . . . . . . . . 85

LIST OF FIGURES xvii

3.22 VBAP Oﬀ center ITD displaced HATS . . . . . . . . . . . . . . 85

3.23 VBAP ITD considering real sound sources only . . . . . . . . . 86

3.24 Ambisonics ITD as a function of source angle . . . . . . . . . . 87

3.25 ILD (VBAP and Ambisonics) in oﬀ-center setups . . . . . . . . 88

3.26 ILD centered setup and oﬀ-center VBAP setups . . . . . . . . . 89

3.27 Diﬀerences in the ILD between centered and oﬀ-center (25 cm)

in VBAP setups . . . . . . . . . . . . . . . . . . . . . . . . . . . 90

3.28 Diﬀerences in the ILD between centered and oﬀ-center (50 cm)

in VBAP setups . . . . . . . . . . . . . . . . . . . . . . . . . . . 90

3.29 Diﬀerences in the ILD between centered and oﬀ-center (75 cm)

in VBAP setups . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

3.30 ILD centered setup and oﬀ-center Ambisonics setups . . . . . . 92

4.1 Auralization procedure implemented to create mixed audible

HINT sentences with 4 spatially separated talkers at the sides

and back (maskers) and one target in front. . . . . . . . . . . . 104

4.2 Spatial setup of the experiment . . . . . . . . . . . . . . . . . . 105

4.3 Eriksholm Anechoic Room: Reverberation Time . . . . . . . . . 105

4.4 Eriksholm Anechoic Room: Background noise . . . . . . . . . . 106

4.5 Experiment setup placed inside anechoic room. . . . . . . . . . . 106

4.6 Overall reverberation time (RT) as a function of receptor (head)

position in the mid-saggital plane re center (0 cm) . . . . . . . . 108

4.7 Sound pressure level Ambisonics virtualized setup . . . . . . . . 109

4.8 Participant positioned to the test. . . . . . . . . . . . . . . . . . 110

4.9 Experiment’s trial design . . . . . . . . . . . . . . . . . . . . . . 111

4.10 Graphic User Interface . . . . . . . . . . . . . . . . . . . . . . . 112

LIST OF FIGURES xviii

4.11 Performance accuracy (word-scoring) . . . . . . . . . . . . . . . 114

4.12 Self reported subjective intelligibility . . . . . . . . . . . . . . . 115

4.13 Self reported subjective eﬀort . . . . . . . . . . . . . . . . . . . 115

4.14 Self reported subjective disengagement . . . . . . . . . . . . . . 115

5.1 Top view. Loudspeakers position on horizontal plane to virtu-

alization with proposed Iceberg method. . . . . . . . . . . . . . 123

5.2 Normalized Ambisonics ﬁrst-order RIR generated via ODEON

software. Left panel depicts the waveform; right panel depicts

the waveform in dB. . . . . . . . . . . . . . . . . . . . . . . . . 125

5.3 Reﬂectogram split into Direct Sound Early and Late Reﬂections. 126

5.4 Iceberg’s processing Block diagram. The Ambisonics RIR is

treated, split, and convolved to an input signal. A virtual audi-

tory scene can be created by playing the multi-channel output

signal with the appropriate setup.. . . . . . . . . . . . . . . . . 129

5.5 Omnidirectional channel of Ambisonics RIR for a simulated room.129

5.6 RIR Ambisonics segments . . . . . . . . . . . . . . . . . . . . . 130

5.7 Example of signal auralized with the Iceberg method . . . . . . 132

5.8 Loudspeaker frequency response comparison . . . . . . . . . . . 135

5.9 Loudspeakers normalized frequency response . . . . . . . . . . . 135

5.10 Loudspeakers normalized frequency response ﬁltered . . . . . . . 136

5.11 BRIR/RIR acquisition ﬂowchart: Iceberg auralization method. . 141

5.12 BRIR measurement setup: B&K HATS and KEMAR positioned

inside the anechoic room. . . . . . . . . . . . . . . . . . . . . . . 141

5.13 Measurement positions (grid) . . . . . . . . . . . . . . . . . . . 142

5.14 RT within Iceberg virtualized environment . . . . . . . . . . . . 144

5.15 Iceberg Centered Interaural Time Diﬀerence . . . . . . . . . . . 145

LIST OF FIGURES xix

5.16 Iceberg Centered Interaural Time Diﬀerence RTs = 0, 0.5 and

1.1 s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146

5.17 Iceberg centered ILD . . . . . . . . . . . . . . . . . . . . . . . . 147

5.18 Iceberg and Real loudspeakers ILDs as a function of azimuth

angle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148

5.19 Iceberg Heatmap Absolute ∆ ILD . . . . . . . . . . . . . . . . 148

5.20 Iceberg: Heatmap Absolute ∆ ILD minus JND . . . . . . . . . . 149

5.21 Iceberg method: Estimated azimuth angle . . . . . . . . . . . . 150

5.22 Iceberg ITD Frontal displacement . . . . . . . . . . . . . . . . . 152

5.23 Iceberg ITD Frontal and lateral displacement . . . . . . . . . . 152

5.24 Delta Interaural Level Diﬀerences RT = 0.0 s . . . . . . . . . . 153

5.25 Delta Interaural Level Diﬀerences RT = 0.5 s . . . . . . . . . . 154

5.26 Delta Interaural Level Diﬀerences RT = 1.1 s . . . . . . . . . . 155

5.27 Estimated (model by May and Kohlrausch [182]) frontal az-

imuth angle at diﬀerent positions inside the loudspeaker ring as

function of the target angle. . . . . . . . . . . . . . . . . . . . . 156

5.28 ITD with second listener present . . . . . . . . . . . . . . . . . 157

5.29 Delta Interaural Level Diﬀerences Centered+Second Listener . . 158

5.30 Estimated localization error with presence of a second listener . 160

5.31 Diﬀerence to target in estimated localization with presence of a

second listener . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161

5.32 Estimated error to RT=0 considering the estimation of real

loudspeakers as basis. . . . . . . . . . . . . . . . . . . . . . . . . 162

5.33 HATS wearing a Oticon Hearing Device (right ear) . . . . . . . 164

5.34 Interaural Time Diﬀerence Iceberg method (aided) . . . . . . . . 165

5.35 Iceberg method ILD (aided condition) . . . . . . . . . . . . . . 166

5.36 Azimuth angle estimation (aided condition) . . . . . . . . . . . 167

5.37 Absolute diﬀerence in estimated azimuth angle (aided condition) 168

5.38 Tukey test to compare means aided condition. . . . . . . . . . . 169

5.39 Iceberg method oﬀ center ITD (aided condition) . . . . . . . . . 170

5.40 Delta Interaural Level Diﬀerences Aided RT = 0.0 s . . . . . . . 172

5.41 Delta Interaural Level Diﬀerences Aided RT = 0.5 s . . . . . . . 172

5.42 Delta Interaural Level Diﬀerences Aided RT = 1.1 s . . . . . . . 173

5.43 Estimated frontal azimuth angle on diﬀerent positions (aided

condition) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174

A.1 ITD as a function of source angle Ambisonics . . . . . . . . . . 225

B.1 Diﬀerences in the ILD Ambisonics, 25 cm . . . . . . . . . . . . . 226

B.2 Diﬀerences in the ILD Ambisonics, 50 cm . . . . . . . . . . . . . 227

B.3 Diﬀerences in the ILD Ambisonics, 75 cm . . . . . . . . . . . . . 227

D.1 Reverberation time (a) Classroom (b) Restaurant . . . . . . . . 231

E.1 Classroom alpha coeﬃcients . . . . . . . . . . . . . . . . . . . . 232

E.2 Restaurant alpha coeﬃcients . . . . . . . . . . . . . . . . . . . . 233

E.3 Anechoic alpha coeﬃcients . . . . . . . . . . . . . . . . . . . . . 233

Nomenclature

General Symbols

Clarity: the ratio between the ﬁrst 50 ms of the RIR and from 50 ms

to the end, Eq. (2.12), page 36.

Clarity: the ratio between ﬁrst 80 ms of the RIR and RIR from 80 ms

to the end, Eq. (2.12), page 36.

Clarity: ratio between ﬁrst RIR 50 ms of a RIR and the complete

RIR, Eq. (2.14), page 37.

Clarity: ratio between ﬁrst 80 ms of a RIR and the the complete

RIR, Eq. (2.14), page 37.

g Gain matrix, page 25.

h(t) Impulse response energy in time domain, Eq. (2.10), page 35.

(t) RIR measured with a pressure gradient microphone, Eq. (2.16), page 38.

(t) Impulse responses collected from the left ear, page 39.

(t) Impulse responses collected from the right ear, page 39.

Vector from center point to channel 1, Eq. (2.3), page 25.

Vector from center point to channel 2, Eq. (2.3), page 25.

Speaker Position Matrix (Channels), page 25.

m Ambisonics components order, Eq. (2.9), page 30.

N number of necessary sources to Ambisonics reproduction, Eq. (2.9),

page 30.

p Vector from center point to virtual font, Eq. (2.3), page 25.

xxi

p p-value for the t-statistic of the hypothesis test that the correspond-

ing coeﬃcient is equal to zero or not., page 114.

(t) Bandpassed left impulse response, Eq. (3.7), page 71.

(t) Bandpassed right impulse response, Eq. (3.7), page 71.

(t) Impulse response at the entrance of the left ear canal, Eq. (3.5),

page 70.

(t) Impulse response at the entrance of the right ear canal, Eq. (3.5),

page 70.

RT Reverberation time, page 34.

Reverberation time, page 34.

s(t) Arbitrary sound source signal, page 29.

(t) Time signal recorded with the set microphone and the loudspeaker

l, page 66.

SE Standard error of the coeﬃcients, page 114.

Reverberation Time (T

) extrapolated from 25 dB of energy decay,

Eq. (2.10), page 35.

Reverberation Time (T

) extrapolated from 35 dB of energy decay,

Eq. (2.10), page 35.

t Time, page 12.

Center Time, Eq. (2.15), page 37.

Reverberation Time, Eq. (2.10), page 34.

v(t)

1kHz

Sinusoidal 1 k Hz signal recorded from the calibrator in VFS, Eq. (3.2),

page 66.

V Volume of the room, Eq. (2.10), page 34.

(t) Calibrator signal recorded in the left ear, Eq. (3.1), page 65.

(t) Calibrator signal recorded in the right ear, Eq. (3.1), page 65.

Greek Symbols

l,rms

Calibration factor for the left ear, Eq. (3.1), page 65.

xxii

r,rms

Calibration factor for the right ear, Eq. (3.1), page 65.

¯α Averaged Absorption Coeﬃcient, Eq. (2.10), page 34.

Level factor to the loudspeaker l, Eq. (3.3), page 66.

ω Angular frequency, page 12.

ϕ Elevation angle related to the ears axis of the listener, page 29.

θ Azimuthal angle related to the ears axis of the listener, page 29.

Mathematical Operators and Conventions

β Fixed-eﬀects regression coeﬃcient, page 114.

e Exponential function, where e

(1)

≈ 2, 7182, page 12.

Integral, page 12.

√

−1, imaginary operator, page 12.

τ Time delay, Eq. (2.18), page 39.

t-statistic for each coeﬃcient to test the null hypothesis, page 114.

(θ, ϕ) Spherical harmonics function of order n and degree m, Eq. (2.7),

page 29.

Equivalent continuous sound level, page 103.

max() Function that returns the element with the maximum value for a

sequence of numbers, or for a vector, Eq. (2.18), page 39.

RMS() Root mean square, Eq. (3.2), page 66.

Acronyms and Abbreviations

2D Two-dimensions in space, page 24.

3D Three-dimensions in space, page 24.

vs. From Latin Versus is the past participle of vertare. which means

“against” and “as opposed or compared to., page 81.

AD/DA Analog-to-Digital Digital-to-Analog converter, page 59.

AR Augmented reality, page 27.

xxiii

ASW Apparent Source Width, page 38.

BRIR Binaural room impulse response, page 15.

CTC Cross-talk cancellation, page 23.

dB HL Hearing Loss in decibels, page 102.

DBAP Distance-Based Amplitude Panning, page 47.

DS Direct sound, page 15.

EcoEG Combination study number 3: Eco (Reverberation/Ecological) and

EEG, page 97.

EEG Electroencephalogram, page 50.

FFRs Brainstem frequency responses, page 50.

FFT Fast Fourier transform, page 70.

FIR Finite Impulse Response, page 71.

HATS Head and torso simulator, page 60.

HC B& K 4128 HATS at center position, page 79.

K+X

B& K 4128 HATS at center position and KEMAR at X cm to the

left, page 80.

HC K-

B& K 4128 HATS at center position and KEMAR at X cm to the

right, page 79.

HEAR-

ECO

Innovative Hearing Aid Research – Ecological Conditions and Out-

come Measures, page 97.

HINT Hearing in Noise Test, page 103.

HOA Higher Order Ambisonics, page 31.

HRTF Head-Related Transfer Function, page 12.

IACC Interaural Cross-Correlation Coeﬃcient, page 39.

IACF Interaural cross-correlation function, Eq. (3.5), page 70.

ILD Interaural Level Diﬀerence, page 10.

IPD Interaural Phase Diﬀerence, page 10.

xxiv

ITD Interaural Time Diﬀerence, page 10.

ITF Interaural Transfer Function, page 14.

JND Just noticeable diﬀerence, page 93.

KEMAR Knowles Electronics Manikin for Acoustic Research, page 60.

LEF Lateral Energy Fraction, Eq. (2.16), page 38.

LEV Listener Envelopment , page 39.

LG Lateral Strength, Eq. (2.17), page 39.

LMM Linear Mixed-eﬀect Model, page 113.

LPF Low-pass ﬁlter, page 73.

LTI Linear and Time-Invariant System, page 33.

MDAP Multiple-Direction Amplitude Panning, page 47.

MOA Mixed Order Ambisonics, page 52.

MTF Monaural Transfer Function, page 13.

NSP Nearest Speaker, page 52.

PLE Perceptual Localization Error, page 41.

PTA4 Four bands pure tone audiometry, page 102.

RIRs Room impulse response, page 15.

SH Spherical Harmonics, page 28.

SPL Sound pressure level, page 35.

SRT Speech Reception Threshold, page 52.

VBAP Vector-Based Amplitude Panning, page 23.

VBIP Vector-Based Intensity Panning, page 41.

VFS Volts full scale, page 66.

VSE Virtual Sound Environment, page 18.

W Omnidirectional channel, Eq. (2.9), page 29.

WFS Wave Field Synthesis, page 31.

xxv

LIST OF FIGURES xxvi

X Bi-directional pattern channel towards the source, Eq. (2.9), page 29.

Y Bi-directional pattern channel perpendicular to the source in az-

imuth, Eq. (2.9), page 29.

Z Bi-directional pattern channel perpendicular to the source in eleva-

tion, Eq. (2.9), page 29.

Chapter 1

Introduction

Individuals with normal hearing often can eﬀortlessly comprehend complex

listening scenarios involving multiple sound sources, background noise, and

echoes [226]. However, those with hearing loss may ﬁnd these situations par-

ticularly challenging [273, 289, 304, 317]. These environments are commonly

encountered in daily life, particularly during social events. They can negatively

impact the communication abilities of individuals with hearing loss [137, 260].

The diﬃculties associated with understanding complex listening scenarios can

be a signiﬁcant barrier for individuals with hearing loss, leading to reduced

participation in social activities [16, 63, 119].

1.1 Motivations

Several hearing research laboratories worldwide are developing systems to re-

alistically simulate challenging scenarios through virtualization to better un-

derstand and help with these everyday challenges in people’s lives [41, 79,

102, 116, 118, 160, 161, 188, 195, 218–220, 259, 272, 298] The virtualization

of sound sources is a powerful tool for auditory research capable of achieving

Chapter 1. Aims and Scope 2

a high level of detail, but current methods use expensive, expansive technol-

ogy [293]. In this work, a new auralization method has been developed to

achieve sound spatialization with a reduction in the technological hardware

requirement, making virtualization at the clinic level possible.

1.2 Aims and Scope

Overall the objective of the research was to investigate parameters of sound

virtualization methods related to its localization accuracy, especially the per-

ceptually based ones [39], in their optimal but also in challenging conditions.

Furthermore, an auralization method oriented to a smaller setup to reduce the

hardware requirements is proposed.

The speciﬁc objectives were:

• To investigate spatial distortions through binaural cue diﬀerences in two

well-known virtualization setups (Vector-Based Amplitude Panning and

Ambisonics (VBAP)).

• To investigate the inﬂuence of a second listener inside the sound ﬁeld

(VBAP and Ambisonics).

• To evaluate the feasibility of a speech-in-noise test within Ambisonics

virtualized reverberant rooms.

• To study the relation between reverberation, signal-to-noise ratio (SNR),

and listening eﬀort in environments virtualized in ﬁrst-order Ambisonics.

• To investigate the binaural cues, objective level and reverberation time

for a new auralization method utilizing four loudspeakers.

Chapter 1. Contributions 3

• To investigate the inﬂuence of hearing aids on binaural cues and objec-

tive parameters within virtualized scenes utilizing the new auralization

method with an appropriate setup.

The main objective of this research was to examine various parameters of sound

virtualization methods related to their localization accuracy, with a focus on

perceptually-based methods [39], in optimal and challenging conditions. Addi-

tionally, a new auralization method was proposed for a smaller setup to reduce

hardware requirements. The speciﬁc goals of the research included:

• Examining spatial distortions through diﬀerences in binaural cues in two

well-known virtualization setups (Vector-Based Amplitude Panning and

Ambisonics (VBAP)).

• Evaluating a second listener’s impact within the sound ﬁeld (VBAP and

Ambisonics).

• Assessing the feasibility of a speech-in-noise test within Ambisonics vir-

tualized reverberant rooms.

• Investigating the relationship between reverberation, signal-to-noise ratio

(SNR), and listening eﬀort in environments virtualized using ﬁrst-order

Ambisonics.

• Using four loudspeakers, propose an auralization method, measure it,

analyze objective parameters against existent methods.

• Test and analyze the inﬂuence of acquiring signals with hearing aids

microphones on virtualized scenes using the new auralization method

with a four-loudspeaker virtualization setup.

Chapter 1. Contributions 4

1.3 Contributions

The main contribution of this research to the scientiﬁc ﬁeld of auditory per-

ception is the development of a new auralization method that addresses the

current gap in the virtualization of sound sources using a small number of

loudspeakers. Speciﬁcally, this method aims to achieve both good localization

accuracy and a high level of immersion simultaneously, which has been a chal-

lenge in previous approaches. Furthermore, the proposed method combines

existing techniques. It can be implemented using readily available hardware,

requiring a minimum of four loudspeakers. This technology makes it more

accessible for audiologists and researchers to create realistic listening scenarios

for patients and participants while reducing the technical resources required for

implementation. Overall, this work represents a valuable contribution to the

ﬁeld of auditory perception and has the potential to advance the understanding

of spatial hearing and the development of eﬀective hearing solutions.

1.4 Organization of the Thesis

In Chapter 2, a review examines previous work carried out in several dif-

ferent areas concerning virtualization and the auralization of sound sources.

The chapter starts with an overview of the basic concepts of human sound

perception. Next, virtual acoustics are explored, reviewing the generation of

virtual acoustic environments using diﬀerent rendering paradigms and meth-

ods. In addition, relevant room acoustics concepts and objective parameters,

and their relation to hearing perception, are described. Finally, the review

considers auralization and virtualization as applied to auditory research. This

review stresses the importance of virtual sound sources for greater realism

and ecological validity in auditory research and the challenges of adequately

Chapter 1. Organization of the Thesis 5

creating a virtual environment focused on auditory research.

Chapter 3 presents an investigation of binaural cue distortions in imperfect

setups. First, the methods are described, including the complete auralization

of signals using two diﬀerent methods and the system’s calibration. The inves-

tigation ﬁrst compares both auralization methods through the same calibrated

virtualization setup in terms of spatial distortions. Then the spatial cues are

examined with the addition of a second listener to the virtualized sound ﬁeld.

Both investigations are performed with the primary listener on and oﬀ-center.

In Chapter 4, a behavioral study examines subjective eﬀort within virtualized

sound scenarios. As the study was part of a collaborative project, only one

auralization method was selected, ﬁrst-order Ambisonics. The aim was to ex-

amine how SNR and reverberation combine to aﬀect eﬀort in a speech-in-noise

task. Also, the feasibility of using ﬁrst-order Ambisonics was examined. How-

ever, the sound sources were well separated in space, and localization accuracy

was not a factor. An important aspect of the study was an auralization issue

involving head movement observed during pilot data collection. This issue

led to a solution that allowed the study to continue. The results veriﬁed the

relationships between subjective eﬀort and acoustic demand. Furthermore,

this issue led to the further investigation of the eﬀect of oﬀ-center listening,

considered in both Chapter 3 and Chapter 5.

In Chapter 5, a hybrid method of auralization is proposed combining the meth-

ods examined and used in previous chapters: VBAP and Ambisonics. This

method was designed to allow auralized signals to be virtualized in a small

reproduction system, thus providing better accessibility to research within the

virtualized sound ﬁeld in clinics and research centers that do not have a size-

able acoustic apparatus. The hybrid auralization method aims to unite the

strengths of both techniques: localization by VBAP and immersion by Am-

Chapter 1. Organization of the Thesis 6

bisonics. Both of these psychoacoustic strengths are related to the room’s im-

pulse response. The hybrid method convolves the desired signal with distinct

parts of an Ambisonics-format impulse response that characterizes the desired

environment. The potential for generating auralizations for a reproduction

system with at least four loudspeakers is demonstrated. The virtualization

system was tested with three diﬀerent scenarios. Parameters relevant to the

perception of a scene, such as reverberation time, sound pressure level, and

binaural cues, were evaluated in diﬀerent positions within the speaker arrange-

ment. The eﬀects of a second participant inside the ring were also investigated.

The evaluated parameters were as expected, with the listener in the system’s

center (sweet spot). However, deviations and issues at speciﬁc presentation

angles were identiﬁed that could be improved in future implementations. Such

errors also need to be further investigated as to their inﬂuence on the subjec-

tive perception of the scenario, which was not performed due to the COVID-19

pandemic. An alternative robustness assessment was performed oﬄine, exam-

ining the localization accuracy with a model proposed by May et al. [182] The

method also proved eﬀective for tests with hearing aids for listeners positioned

in the center of the speaker arrangement. However, the method performance

considering hearing instruments with compression algorithms and advanced

signal processing still needs to be veriﬁed.

Chapter 6 presents a general discussion of the feasibility of applying tests

using the proposed method and an overview of the processes. In addition, the

relevant contributions of the work are presented, as are the limitations and the

suggestions for further improvements.

Chapter 2

Literature Review

2.1 Introduction

The ﬁeld of audiology is concerned with the study of hearing and hearing dis-

orders, as well as the assessment and rehabilitation of individuals with hearing

loss [110]. In this review chapter, we will explore various topics related to

human binaural hearing, spatial sound, and virtual acoustics to provide a

comprehensive overview of the current state of knowledge in these ﬁelds and

highlight their important contributions to our understanding of hearing and

auditory perception. First, we will delve into the intricacies of human binaural

hearing. Next, we will examine the concepts of spatial hearing, including the

various binaural and monoaural cues that contribute to our ability to local-

ize sound in space. We will also explore the head-related transfer function,

which describes the way that sounds are ﬁltered as they travel from their

source to the ear drum, as well as the subjective aspects of audible reﬂections.

Next, we will turn our attention to spatial sound and virtual acoustics. We

will discuss the virtualization of sound, including the various methods used

to achieve this, such as auralization and virtual sound reproduction. We will

Chapter 2. Human Binaural Hearing 8

also examine the diﬀerent auralization paradigms used in auditory research,

including binaural, panorama, vector-based amplitude panning, ambisonics,

and sound ﬁeld synthesis. We will then examine the role of room acoustics in

virtualization, and auditory research, including the various parameters, used

to describe room acoustics, such as reverberation time, clarity and deﬁnition,

center time, and parameters related to spatiality. Finally, we will explore the

use of loudspeaker-based virtualization in auditory research, including hybrid

methods and sound source localization, as well as the assessment of listening

eﬀort.

2.2 Human Binaural Hearing

The engineering side of the listening process can be simpliﬁed modeled through

two input blocks separated in space [92]. These inputs, frequency, and level

are limited and are followed by a signal processing chain that relates the

medium transformations for the wave propagation from air to ﬂuid and elec-

trical pulses [315].

Although this block modeling can be reasonably accurate for educational pur-

poses, it falls short of capturing the true eﬀect and importance of listening

on our essence as human beings. The ability to feel and interpret the world

through the sense of hearing, and to attribute meaning to sound events, enables

humans to enrich their tangible world [56, 244]. For instance, a characteristic

sound can evoke memories or trigger an alert [128]. A piece of music can bring

tears to one’s eyes or persuade someone to purchase more cereal [13, 114]. A

person’s voice can activate certain facial nerves, turning hidden teeth into a

smile. These are some of the reasons why researchers and clinicians dedicate

their lives to understanding the transformation of sound events into auditory

events, with a scientiﬁc dedication focused on creating solutions and opening

Chapter 2. Human Binaural Hearing 9

opportunities for more people to experience the sound they love and deserve -

a dedication focused on people and their needs.

As the auditory system comprises two sensors, normal-hearing listeners can

experience the beneﬁts of comparing sounds autonomously, relating them to

the space around them [21]. This constant signal comparison is the main

principle of binaural hearing, where the diﬀerences between these sounds allow

for the identiﬁcation of the direction of a sound event, as well as the sensation

of sound spatiality [9, 40]. Usually, these signals are assumed to be part of a

linear and time-invariant system, which helps to study how humans interpret

the information present in the diﬀerent signals across the time and frequency

domains. However, this assumption of linearity can fail when analyzing fast

sound sources, reﬂective surfaces, or sound propagating through disturbed

air [200, 255]. Nonetheless, the advantages of quantifying and capturing the

eﬀect have led to signiﬁcant progress in hearing sciences.

2.2.1 Spatial Hearing Concepts

Identifying the direction of incidence of a sound source based on the audible

waves received by the listener is deﬁned as an act or process of human sound

localization [285]. For research in acoustics, it is relevant to acknowledge that

the receiver is, in general, a human being. The human hearing mechanism’s

main anatomical characteristic is the binaural system. There are two signal

reception points (external ears positioned on opposite sides of the head). Al-

beit, the whole set (torso, head, hearing pavilions) can also modify the signal

that reaches the two tympanic membranes at some extent [153, 216]. Hu-

man binaural hearing and associated eﬀects have been extensively reported by

Blauert [38].

Chapter 2. Human Binaural Hearing 10

In addition to analyzing sound sources’ spatial location, the central auditory

system extracts real-time information from the sound signals related to the

acoustic environment, such as geometry and physical properties [153]. An-

other beneﬁt is the possibility of separating and interpreting combined sounds,

especially from sources in diﬀerent directions [170, 242].

2.2.2 Binaural cues

The sound propagation speed in the air can be assumed to be ﬁnite and

approximately constant, considering it as an approximately non-dispersive

medium [18]. Thus, when the incidence is not directly frontal or rear, the

wavefront travels through diﬀerent paths to the ears, reaching them at diﬀer-

ent times. The time interval between that a sound takes to arrive on both

ears is commonly expressed in the literature as Interaural Time Diﬀerence

(ITD) [39]. It is crucial cue for sound source localization in low-frequency

sounds [39, 153, 242]. Moreover, it is considered the primary localization

cue [306]. For continuous pure tone signals and other periodic signals, the

ITD can be expressed as the time Interaural Phase Diﬀerence (IPD) [285].

On the other hand, most mammals’ high-frequency sound source localization

is based on a comparative analysis of sound energy in each ear’s frequency

bands, the Interaural Level Diﬀerence (ILD). The named duplex theory sur-

mises ITD cues as the basis to sound localization of low-frequency and ILD

cues to high-frequency. The authorship of this is assigned to Lord Rayleigh at

the beginning of the last century [246]. These binaural cues are related to the

azimuthal position. However, they do not present the same success explaining

the localization on elevated positions [37, 250]. An ambiguity in binaural cues

caused by head symmetry and referred to as the cone of confusion [296] can

create diﬃculties to a correct sound source localization. The cone of confusion

is the imaginary cone extended sideways from each ear where sound source

Chapter 2. Human Binaural Hearing 11

locations will create the same interaural diﬀerences (see Figure 2.1).

Figure 2.1: Two-dimensional representation of the cone of confusion.

Head movements are essential for resolving the ambiguous cues from sound

sources located on the cone of confusion. As the person moves their head,

they change the reference and the incidence angle helping them to solve the

duality. This change is reﬂected in the cues associated with directional sound

ﬁltering caused by the human body’s reﬂection, absorption, and diﬀraction.

2.2.3 Monaural cues

Monoaural cues are related to spatial impression, especially in the localization

of elevated sound sources. These cues give, to some extent, some limited but

crucial localization abilities to people with unilateral hearing loss [72, 307].

This type of cue is centered on instant level comparison and frequency changes.

As the level of a continuous enough sound source changes, the approximation

or distancing of that source can be estimated. Furthermore, when there are

head movements that shape the frequency content, the disturbance, mainly

the pinnae provide, can beneﬁt the listener to learn the position of a sound

source [129, 292]. In addition, the importance of the previous knowledge of

the sound to the deconvolution process is also investigated, revealing mixed

results [307].

Chapter 2. Human Binaural Hearing 12

2.2.4 Head-related transfer function

The Head-Related Transfer Function (HRTF) describes the directional ﬁltering

of incoming sound due to human body parts such as the head and pinnae [189].

The free-ﬁeld HRTF can be expressed as the division of the impulse responses

in the frequency domain measured at the entrance to the ear canal and the

center of the head but with the head absent [108] (see Figure 2.2).

HRTFs depend on the direction of incidence of the sound and are generally

measured for some discrete incidence directions. Mathematical models can

also generate individualized HRTFs based on anthropometric measures [52] or

through geometric generalization [70].

Figure 2.2: A descriptive deﬁnition of the measured free-ﬁeld HRTF for a given

angle.

The referential system related to the head can be seen in Figure 2.3, where

β is the elevation angle in the midplane, and ϕ is the angle deﬁned in the

horizontal plane.

Chapter 2. Human Binaural Hearing 13

Figure 2.3: Polar coordinate system related to head incidence angles, adapted from

Portela [240].

Suppose the distance to the sound source exceeds 3 meters. In that case, it can

be considered approximately a plane wave, thus making the previous HRTFs

almost independent of the distance to the sound source [38]. Blauert [39] also

explain two other types of HRTF, namely:

• Monaural Transfer Function (MTF): relates the sound pressure, at a

measurement point in the ear canal, from a sound source at any position

to a sound pressure measured at the same point, with a sound source at

a reference position (ϕ = 0 and β = 0). MTF is given by

MTF =





r,ϕ,β,f





ϕ=0

◦

,β=0

◦

, (2.1)

where p

it can be p

, p

or p

– p

sound pressure in the center of the head position with the listener

Chapter 2. Human Binaural Hearing 14

absent;

– p

sound pressure at the entrance of the occluded ear canal;

– p

sound pressure at the entrance to the ear canal;

– p

: eardrum sound pressure.

• Interaural Transfer Function (ITF): relates the sound pressures at corre-

sponding measurement points in the two auditory canals. The reference

pressure will then be the ear that is directed towards the sound source.

The ITF can be obtained through

ITF =

i Opposite side of the source

i Side facing the source

. (2.2)

More considerable variations are seen above 200 Hz in HRTFs [293] because

the head, torso, and shoulders begin to signiﬁcantly interfere in frequencies

up to approximately 1.5 kHz (mid frequencies). In addition, the pinna and

the cavum conchae (space inside the most inferior part of the helix cross; it

forms the vestibule that leads into the external acoustic meatus [270]) distort

frequencies greater than 2 kHz.

HRTF measurements vary from person to person, as seen in Figure 2.4, where

TS 1, TS 2, TS 3, and TS 4. represent HRTFs of diﬀerent people. When

recording using mannequins or diﬀerent people’s ear canals (non-individualized

HRTFs), the reproduction precision in terms of spatial location and realism

tends to be diminished [51, 178]. This poorer precision is because the transfer

function will diﬀer for each individual, especially at high frequencies [155].

This dependence is related to the wavelength and the singular irregularity of

the ear canal of each human being [38].

Chapter 2. Human Binaural Hearing 15

Figure 2.4: Head-related transfer functions of four human test participants, frontal

incidence, from Vorl¨ander [293].

Binaural Impulse Response A Binaural Room Impulse Response (BRIR)

results from a measurement of the response of a room to excitation from an

(ideally) impulsive sound [183]. The BRIRs are composed of a sequence of

sounds. Parameters like the magnitude and the decay rate, the phase, and

time distribution are the key to understanding how a BRIR can audibly char-

acterize a room to a human perception [167]. Albeit the air contains a small

portion of Co2 that is dispersive, sound propagation velocity can be considered

homogeneous in the air (non-dispersive medium) [312] for the Room Impulse

Responses (RIRs). The ﬁrst sound from a source that reaches a receptor inside

the room travels a smaller distance, and it is called direct sound (DS). Usually,

the following sounds result from reﬂections that travel a longer path, losing

energy on each interaction and resulting in an exponential decay of magnitude.

The BRIR is proposed to collect the room information as a regular Impulse

Response, although having two sensors separated as the typical human head.

Nowadays, BRIR can be recorded with small microphones placed in the ear

canal of a person or utilizing microphones placed in mannequins [197].

A BRIR is the auditory time representation of a set source-receptor deﬁned

by its position, orientation, acoustic properties as directionality of the sound

Chapter 2. Human Binaural Hearing 16

source, as well as from the physical elements within the environment [38, 108].

The convolution of BRIR with audio signals is a feasible task for modern

computation, which allows the creation and manipulation of sounds even in

real-time applications [62, 217]. Thus, it is possible to impose spatial and

reverberant characteristics of diﬀerent spaces to a given sound [109].

2.2.5 Subjective aspects of an audible reﬂection

The impulse response is composed of the direct sound followed by a series of

reﬂections (initial and later reﬂections) [45, 165]. Essential knowledge on how

the human auditory system processes the spectral and spatial information

contained in the impulse response has been obtained through studies with

simulated acoustic ﬁelds. [6, 17, 93, 125, 141, 174, 176, 188, 193, 257, 305,

321]. The results of Barron’s experiments, depicted in Figure 2.5, involved

the reproduction of both a direct sound and a lateral reﬂection. These two

auditory stimuli were manipulated in terms of their time delay and relative

amplitude, with the goal of eliciting subjective impressions correlated to these

factors. By varying the time between the direct sound and the reﬂection, as

well as the relative amplitude of these stimuli, it was possible to understand

better how these characteristics impact the overall auditory experience.

Figure 2.5: Audible eﬀects of a single reﬂection arriving from the side (adapt from

Rossing [254]).

Chapter 2. Spatial Sound & Virtual Acoustics 17

The audibility threshold curve indicates that the reﬂection will be inaudible if

the delay or the relative level is minimal. The reﬂection’s subjective eﬀect also

depends on the direction of incidence of the sound source in the horizontal and

vertical plane. It is possible to note that for delays of up to 10 milliseconds,

the relative diﬀerence in level must be at least -20 dB for the reﬂection to be

noticeable.

The echo eﬀect is typically observed in delays of more than 50 milliseconds,

being an acoustic repetition with a high relative level—approximately the same

energy as the direct sound. The coloring eﬀect is associated with the signiﬁcant

change in the spectrum caused by the constructive and destructive interference

of the superposition of sound waves.

The image change happens when there are reﬂections with relative levels higher

than direct sound or minimal delays. In this case, the subjective perception is

that the sound source has a diﬀerent position in space than the visual system

perceives.

2.3 Spatial Sound & Virtual Acoustics

The sound perceived by humans is identiﬁed and classiﬁed based on physical

properties, such as intensity and frequency [242]. Human beings are equipped

with two ears (two highly eﬃcient sound sensors), enabling a real-time compar-

ison of these properties between the captured sound signals [9]. The sounds

and the dynamic interaction between sound sources, their positions, move-

ments, and the physical interaction of the generated sound waves with the

environment can be perceived by normal-hearing people, providing what is

called spatial awareness [153]. That auditory spatial awareness includes the

localization of the sound source, estimation of distance, and estimation of the

Chapter 2. Spatial Sound & Virtual Acoustics 18

size of the surrounding space [38, 305]. A person with hearing loss may lose

this ability partially or entirely; the spatial awareness is also tied to the lis-

tener’s experience with the sound and the environment, motivation, or fatigue

level [54, 304].

In the ﬁeld of virtual acoustics, the ultimate goal is to generate a sound event

that elicits a desired auditory sensation, creating a Virtual Sound Environment

(VSE) [293]. In order to achieve this, it is necessary to synthesize or record

the acoustic properties of the target scene and subsequently reproduce them

in a manner that accurately reﬂects the original acoustic conditions [97]. This

involves a careful consideration of the various factors that contribute to the

overall auditory experience, including the spectral and spatial characteristics

of the sound. By accurately recreating these properties, it is possible to create

a highly immersive and realistic VSE that eﬀectively conveys the intended

auditory experience to the listener [196, 213, 293].

2.3.1 Virtualization

Nowadays, it is possible to create audio ﬁles containing information about

sound properties related to a speciﬁc space [293]. For example, it is possible to

encode information about the source and receptor position, the transmission

path, reﬂections on surfaces, and the amount of energy absorbed and scattered

(e.g., Odeon [59], a commercially available acoustical software). The sound

ﬁeld properties can be simulated, synthesized, or recorded in-situ [113, 293].

These signals can be encoded and reproduced correctly in various reproduction

systems [122, 161]. The creation of ﬁles that can be reproduced containing

such information is called auralization. As diﬀerent interpretations of the

terms occur in literature, in this thesis, the virtualization process is considered

to encompass the auralization and the reproduction of a sound (recorded,

Chapter 2. Spatial Sound & Virtual Acoustics 19

simulated, or synthesized) that includes spatial properties.

2.3.1.1 Auralization

Auralization is a relatively recent procedure. The ﬁrst studies were conducted

in 1929; Spand¨ock and colleagues tried to process signals measured in a scale-

created room. After that, in 1934, Spand¨ock [280] succeeded in the ﬁrst aural-

ization, in the analogical way, using ultrasonic signals of scale models recorded

in magnetic tapes. In 1962 Schroeder [263] incorporated the computing pro-

cess into the auralization. In 1968 Krokstad [146] developed the ﬁrst acoustic

room simulation software. The term auralization was introduced in the litera-

ture by Kleiner in 1993: “Auralization is the process of rendering audible, by

physical or mathematical modeling, the soundﬁeld of a source in a space, in

such a way as to simulate the binaural listening experience at a given position

in the modeled space.” (Kleiner [138])

In his book titled Auralization, published in 2008, Vorl¨ander deﬁned: “Aural-

ization is the technique of creating audible sound ﬁles from numerical (simu-

lated, measured, or synthesized) data.” (Vorl¨ander [293])

In this work, auralization is understood as a technique to create ﬁles that can

be executed as perceivable sounds. An auralization method describes the tech-

nique; it can involve one or more auralization techniques. These sounds can

then be virtualized (reproduced) via loudspeakers or headphones and provide

audible information about a speciﬁc acoustical scene in a deﬁned space, fol-

lowing Vorlander’s deﬁnition. That was chosen to encourage the separation of

the process as an auralized sound ﬁle can contain information that allows it to

be decoded in diﬀerent reproduction systems [320].

Auralization is consolidated in architectural acoustics [45, 148, 165, 254], and

Chapter 2. Spatial Sound & Virtual Acoustics 20

it is also emerging in environmental acoustics [19, 68, 69, 139, 162, 231, 232].

This technique allows a piece of audible information to be easily accessed and

understood. It is also an integral part of the entertainment industry in games,

movies, and virtual or mixed reality [320]. Knowing an environment’s acoustic

properties allows it to manipulate or add synthesized or recorded elements,

leading the receiver to the desired auditory impression, including the sound’s

spatial distribution [62]. This process is also used in hearing research, allowing

researchers to introduce more ecologically valid sound scenarios to their study

(See Section 2.3.4).

Sound spatiality, or the perception of sound waves arriving from various direc-

tions and the ability to locate them in space, is a crucial aspect of the auditory

experience [40]. Auralization, which is analogous to visualization, involves the

representation of sound ﬁelds and sources, the simulation of sound propaga-

tion, and the strategy to decode in the spatial reproduction setup [293]. That

is typically achieved through tri-dimensional computer models and digital sig-

nal processing techniques, which are applied to generate auralizations that can

be reproduced via acoustic transducers [293].

The modeling paradigm used to create the spatial sensation can be percep-

tually or physically based [39, 106, 164, 276]. Multiple dimensions inﬂu-

ence sound perception; the type of generation of the sound, the wind direc-

tion, the temperature, the source and the receptor movement, space (size,

shape, and content), receptor’s spatial sensitivity, and source directivity are

some examples. That implies the importance of physical eﬀects as Doppler

shifts [96, 284, 293]. Furthermore, the review of room acoustic and psychoa-

coustics elements (See Section 2.3.3) corroborates the auralization modeling

procedure’s understanding.

Chapter 2. Spatial Sound & Virtual Acoustics 21

2.3.1.2 Reproduction

Sound signals containing acoustic characteristics of a space can be reproduced

either with binaural techniques (headphones or loudspeakers) or with multiple

loudspeakers (multichannel techniques) [293]. Moreover, an acoustic model

for a space can be analytically or numerically implemented, having a series

of competent algorithms and commercial software and tools available [49].

With that, it is also possible to measure micro and macro acoustic properties

for materials in a laboratory or in-situ [206] and access databases of various

coeﬃcients and indexes to an extended catalog of materials [50, 71, 158, 266].

On the reproduction end of the virtualization process, factors such as fre-

quency and level calibration, signal processing, and the frequency response of

the hardware can signiﬁcantly impact the accuracy of the ﬁnal sound (e.g., the

orientation/correction of the microphone when calibrating the system [274]).

Depending on the chosen paradigm, a lack of attention to these details may

disrupt an accurate description of the sound ﬁeld, sound event, or sound sen-

sation [214, 282, 283, 320]. Additionally, the quality of the stimuli may be

compromised depending on the chosen reproduction technique, which is often

tied to the hardware available [77, 166, 275, 276]. That can lead to undesired

eﬀects on the level of immersion and problems with the accuracy of sound

localization and identiﬁcation (e.g., source width, source separation, sound

pressure level, and coloration and spatial confusion eﬀects [97]). The process

of building a VSE is called sound virtualization, which involves both the aural-

ization and reproduction stages to create audible sound from a ﬁle. The main

technical approaches or paradigms for reproducing auralized sound are Binau-

ral, Panorama, and Sound Field Synthesis (Section 2.3.2). These paradigms

can be distinguished by their output, which can be physically or perceptu-

ally motivated. For example, while binaural methods are treated apart, they

can be intrinsically classiﬁed in a physically-motivated paradigm since its suc-

Chapter 2. Spatial Sound & Virtual Acoustics 22

cess relies on reproducing the correct physical signal at a speciﬁc point in the

listener’s auditory system, typically the entrance of the ear canal [106].

2.3.2 Auralization Paradigms

2.3.2.1 Binaural

Binaural hearing, which refers to the ability to perceive sound in a three-

dimensional auditory space, is a fundamental concept in auditory research and

has been extensively studied by researchers such as Blauert [40]. In the con-

text of auralization, the term ”binaural” refers to the speciﬁc paradigm that

aims to reproduce the exact sound pressure level of a sound event at the lis-

tener’s eardrums. That can be achieved through the use of headphones or a

pair of loudspeakers (known as transaural reproduction) [314]. However, when

using distant loudspeakers, it is necessary to consider the interference that can

occur between the sounds coming from each speaker. To mitigate this issue,

techniques such as cross-talk cancellation (CTC) [60, 262] can be employed,

which involve manipulating a set of ﬁlters to cancel out the distortions caused

by the sound from one speaker reaching the other ear. Another form of bin-

aural reproduction involves the use of closer loudspeakers that are nearﬁeld

compensated.

Binaural methods over headphones is commonly applied. It requires no ex-

tensive hardware (in simple setups that do not track the listener’s head), pro-

viding a valid acoustic representation and spatial awareness [293]. A Disad-

vantage of this method can include the accuracy dependence of individualized

HRTF (as each human being have his own slightly diﬀerent anatomic ”ﬁl-

ter set”) [314]. Over headphones also, the movement of the listener’s head

can be disruptive to the immersion [179]. It may require tracking the head’s

Chapter 2. Spatial Sound & Virtual Acoustics 23

movement [11, 115, 252], e.g., when movements are required or allowed in

an experiment. Furthermore, a listener wearing a pair of headphones may

not represent a realistic situation. For example, an experiment with a virtual

auditory environment that represents a regular daily conversation with aged

participants may lose the task’s ecological validity. Also, usually, headphones

prevent the listener from wearing hearing devices. Figure 2.6 illustrates the

main idea behind diﬀerent binaural reproduction setups.

Figure 2.6: Binaural reproduction setups: Headphones, transaural and near-ﬁeld

transaural (Adapted from Kang and Kim [131]).

2.3.2.2 Panorama

The Panorama paradigm encompasses auralization methods focused on deliv-

ering accurate ITDs and ILDs at the listener’s position, also known as stereo-

phonic techniques [106, 276]. The most well-known methods are based on am-

plitude panning [180], including Low-order Ambisonics [91] and Vector-Based

Amplitude Panning (VBAP) [241]. High Order Ambisonics is an extension of

the Ambisonics method, which is not typically considered a panning method

but rather a sound ﬁeld synthesis method (See Section 2.3.2.3). VBAP em-

ploys local panning by rendering sound using pairs or triplets of loudspeakers.

In contrast, Ambisonics uses global panning to produce a single virtual source

using all available loudspeakers [282].

Chapter 2. Spatial Sound & Virtual Acoustics 24

Vector Based Amplitude Panning: The Vector-Based Amplitude Pan-

ning (VBAP) is a ﬁrst-order approximation of the composition of emitted

signals that creates virtual sources [241]. The virtualization process using

VBAP is based on amplitude panning in two dimensions (variation in ampli-

tude between the speakers), which is derived from the Law of Sines and Law of

Tangents (see Benesty et al. [23] for a derivation of these laws). The original

hypothesis of VBAP assumes that the speakers are arranged symmetrically,

equidistant from the listener, and in the same horizontal plane. VBAP does

not limit the number of usable speakers but uses a maximum of 3 simulta-

neously. The speakers are arranged in a reference circle (2D case) or sphere

(3D case), and a limitation of the technique is that virtual sources cannot be

created outside of this region. VBAP is mainly used for the reproduction of

synthetic sounds [180].

The formulation of the VBAP method (from Pulkki [241]) for two dimensions

starts from the stereophonic conﬁguration of two channels (see Figure 2.7).

Reformulated to a vector base, formed by unit length vectors l

= [l

]

and

= [l

]

that point to the speakers and the unit length vector p = [p

]

which points to the virtual source and presents itself as a linear combination

of the vectors l

and l

. The notation

is used here to identify the matrix

transposition.

Chapter 2. Spatial Sound & Virtual Acoustics 25

Figure 2.7: Vector-based amplitude panning: 2D display of sound sources positions

and weights.

Consider the vetor p:

p = g

+ g

, (2.3)

where g

and g

(scalar) are the gain factors to be calculated for positioning

the vector relative to the virtual source. In matrix form, there is

= gL

, (2.4)

where g = [g

], and L

= [l

]

. The gains can be calculated by

g = p

−1

= [p

]













−1

. (2.5)

Chapter 2. Spatial Sound & Virtual Acoustics 26

The formulation is also expanded to 3 dimensions:

p = g

+ g

, (2.6)

and

= gL

123

, (2.7)

where g

, g

, and g

are gain factors, g = [g

], and L

123

= [l

]

The detailed derivation can be found at [241]. The derivation can use triangles

and the three-dimensional system. Figure 2.8 presents an example of the sound

source distribution for virtualization of a virtual source P using VBAP in three

dimensions.

Figure 2.8: Diagram representing the placement of speakers in the VBAP technique

Adapted from [241].

Some factors collaborate so that methods based on Amplitude Panorama are

widely used in virtual audio applications, such as the low computational cost

and ﬂexibility in the speakers’ placement.

Ambisonics: The original Ambisonics auralization method is an amplitude

panning method that diﬀers from the Vector Base Amplitude Panning (VBAP)

Chapter 2. Spatial Sound & Virtual Acoustics 27

method in several ways. While VBAP only uses positive weights to pan

sound across speakers, Ambisonics uses a combination of positive and neg-

ative weights to create a shift in frequency and amplitude. This results in a

more homogeneous sound ﬁeld, albeit with a broader virtual source. Addition-

ally, Ambisonics has all loudspeakers active for any source position. At the

same time, VBAP only activates speciﬁc speakers based on the desired source

position [199].

One of the beneﬁts of Ambisonics is its scalability for reproduction on diﬀerent

loudspeaker arrays and the ability to encode and decode the sound ﬁeld dur-

ing the recording and reproduction process [161]. This versatility is possible

because Ambisonics signals can be directly recorded using an appropriate mi-

crophone array or simulated through numerical acoustic algorithms that model

the directional sensitivity of the microphone array [5, 46, 59]. The signal can

then be decoded and rendered in real time to diﬀerent arrays with various

numbers of loudspeakers. Hence, an Ambisonics decoder is a tool for convert-

ing an Ambisonics representation of a sound ﬁeld into a multichannel audio

format that can be reproduced over a given speaker setup [130, 235, 238]. In

order to reproduce an Ambisonics signal, it must ﬁrst be transformed, or ”de-

coded,” into a format compatible with a speciﬁc speaker conﬁguration. Simple

decoders consist of a frequency-independent weighting matrix [282]. It is also

possible to reproduce the signal via headphones, which can be considered a

speciﬁc speaker setup, by scaling it down to binaural signals [320]. Addi-

tionally, Ambisonics can enhance realism by tracking head movements and

correcting binaural signals utilizing HRTFs as ﬁlters [277]. This feature is par-

ticularly relevant in the recording and broadcasting industry, particularly with

emerging technologies such as augmented reality (AR) [320]. In summary, an

Ambisonics decoder is a tool used to transform an Ambisonics representation

of a sound ﬁeld into a multichannel audio format that can be reproduced over a

given loudspeaker setup, enabling the creation of immersive sound experiences.

Chapter 2. Spatial Sound & Virtual Acoustics 28

According to Schr¨oder [261], decomposition in spherical harmonics (SH) is a

recent analysis and widely used in the modeling of directivity paths. Analo-

gous to a Fourier transform in the frequency domain, SH in the spatial domain

decomposes the signal into spherical functions (in the Fourier transform, the

decomposition is in sine or cosine functions) weighted by the coeﬃcients of the

corresponding harmonic spheres. According to Pollow [239], it is commonly

applied in multi-dimensional domain problems. However, the analytical re-

quirements for cases with few dimensions (two in the case of the sound ﬁeld)

are considerably simpliﬁed. Manipulating the wave equation by separating

variables is an essential tool here.

Appendix C shows the derivation of SH through the separation of variables of

the wave equation in spherical coordinates (Equation C.1).

The solutions to the linear wave equation in spherical coordinates expressed

in the frequency domain (Helmholtz equation) are orthogonal basis functions

(θ, ϕ) where n is the degree and m is the order. These angle-dependent

functions are called spherical harmonics and can represent, for example, a

sound ﬁeld [309]. That is the core assumption to Ambisonics recording and

reproduction. Figure 2.9 depicts SHs up to order N = 2.

Chapter 2. Spatial Sound & Virtual Acoustics 29

Figure 2.9: Spherical Harmonics Y

(θ, ϕ). Rows correspond to orders 0 ≤ n ≤ 2,

columns to degrees −n ≤ m ≤ n (adapted from Pollow [239]).

The four SH weights Y

(θ, ϕ) to encode all the spatial audio information into

a First-Order Ambisonics ﬁle is given by:

(t) = s(t)Y

(θ

, ϕ

) (2.8)

where the s(t) is the source signal in the time domain and Y

(θ

, ϕ

) the

encoding coeﬃcients to the source s(t). Computed as ﬁrst order in the B-

format, the normalized components can be described as [172]:











W = B

= SY

(θ

, ϕ

) = S(0.707)

X = B

= SY

(θ

, ϕ

) = S cos θ

cos ϕ

Y = B

1−1

= SY

1−1

(θ

, ϕ

) = S sin θ

cos ϕ

Z = B

= SY

(θ

, ϕ

) = S sin ϕ

(2.9)

The resulting four-channel signals are the equivalent to an omnidirectional

(W) and three orthogonal bi-directional (commonly called ﬁgure-of-eight) mi-

Chapter 2. Spatial Sound & Virtual Acoustics 30

crophones (X, Y, and Z). The channels can represent the pressure and the

particle velocity of a given sound (See Figure 2.10. It is possible to transcode

and manipulate the generated signal to change its orientation with a matrix

multiplication in signal processing. Also, it is possible to decode the same

encoded signal to a single sound source, headphones, or a multichannel array.

Figure 2.10: B-format components: omnidirectional pressure component W, and

the three velocity components X, Y, Z. Extracted from Rumsey [256].

The limitation of ﬁrst-order Ambisonics is spatial precision since it is only eﬀec-

tive at a point centered within a deﬁned area. This limitation can be overcome

with higher-order components. Adding a set of higher-order components im-

proves the directionality. However, increasing the number of components will

also increase the number of loudspeakers required to play higher-order Am-

bisonics. That means a more accurate sound ﬁeld representation if the order

is increased. The number of channels N for a periphonic Ambisonics of order

m order is N = (m + 1)

for 3D reproduction and N = (2m + 1) for 2D [65].

Chapter 2. Spatial Sound & Virtual Acoustics 31

2.3.2.3 Sound Field Synthesis

The objective of techniques from Sound Field Synthesis remains the same as in

the techniques from the perceptually-motivated paradigm: a spatial sound ﬁeld

reproduction. The perceptually motivated are centered on the psychoacoustic

eﬀects of summing binaural cues that lead the listener to perceive a virtual

source. On the other side, the Sound Field Synthesis techniques rely on the

physical reconstruction of the original/simulated sound ﬁeld to a speciﬁc area.

The main techniques are the extension of Ambisonics reproduction to higher

orders called Higher Order Ambisonics (HOA) and the Wave Field Synthesis

(WFS) [24, 25].

The HOA extends the order of the classical Ambisonics and, therefore, the

number of sound sources arranged in a spherical array. As the Ambisonics

order increases, the perceived sound source direction accuracy also increases,

although requiring more loudspeakers [97]. An important distinction can be

made between the Ambisonics and HOA. Given the truncation possibility in

Ambisonics, the method is treated as a soft transition from a perceptually

based method to a physically based one. Although the HOA utilizes the same

principle as Ambisonics, it is classiﬁed as a sound ﬁeld synthesis paradigm

(physically based) along with WFS. The HOA limitations are reported in the

literature by several studies [26, 27, 64, 73, 236, 299], especially the aliasing in

frequency that leads to pressure errors and the sweet spot size [253].

The WFS formulation relies on Huygen’s principle: a propagating wavefront at

any instant is shaped to the envelope of spherical waves emanating from every

point on the wavefront at the prior instant [281], the principle is illustrated in

Figure 2.11).

Chapter 2. Spatial Sound & Virtual Acoustics 32

Figure 2.11: Illustration of Huygen’s Principle of a propagating wave front.

A conceptual diﬀerence between WFS and HOA is that for HOA, the sound

characteristics are described in a point (or small area) inside the array, while in

WFS, the sound pressure that must be known is on the border of the reproduc-

tion area. A review and a comparison of both methods and their compromises

in terms of spatial aliasing errors and noise ampliﬁcation is presented in [65].

Their ﬁndings indicated similar constraints to both methods. However, they

are both characterized by the requirement of large arrays of loudspeakers. The

HOA has been found to have a higher limit of the size of the center area. In

contrast, the WFS has limitations on the distortion of higher frequencies (alias-

ing), depending on the number of loudspeakers. Regardless, as the scope of

the thesis aims to work with a small number of loudspeakers, they will not be

thoroughly discussed.

2.3.3 Room acoustics

Diﬀerent aspects are taken into account when describing the hearing experi-

ence of a human being in a space, for example, the individuality of auditory

Chapter 2. Spatial Sound & Virtual Acoustics 33

training, familiarity with space, personal preferences, humor, fatigue, culture,

and the spoken language [10, 32, 123, 169, 184, 190, 250]. However, there

are similarities in the expressions used between sample groups. Such an eﬀect

is attributed to the similarity of the auditory-cognitive mechanism of human

beings [40].

In architectural acoustics and room acoustics, studies of sound properties as-

sume that a sound source and a receiver in a given space is a linear and time-

invariant system (LTI) [45]. Thus, a complete LTI characterization to each

source-receiver can be expressed by its impulse response in the time domain

or the transfer function in the frequency domain [45, 293].

2.3.3.1 Room acoustics parameters

Objective parameters are essential in acoustic projects and in compositions

of statistical models that aim to predict the human interpretation of acous-

tic phenomena [254]. Objective parameters derived from the LTI’s impulse

response aim to create metrics that quantify subjective descriptors from nu-

merous experiments [254]. The calculation and measurement of many objective

parameters are described in an appendix to the International Organization for

Standardization (ISO) standard 3382 [127].

2.3.3.2 Reverberation Time

The reverberation time (RT) measures the time it takes for the impulse re-

sponse’s sound pressure level to decrease to one-millionth of its maximum

value, equivalent to a decline of 60 dB; it is also often referred to as RT

. Note that the reverberation time measures how fast the decay of sound

energy occurs and not how long the reverberation lasts in the environment,

Chapter 2. Spatial Sound & Virtual Acoustics 34

depending on the sound source power and the background noise. The RT

was the ﬁrst parameter studied, modeled, and understood, related to several

subjective aspects of the human hearing experience in a room. Today, this is

considered the most critical parameter, although it is not enough to describe

human perception completely. Wallace C. Sabine [258] initially described it

through mathematical relations obtained by an empirical method, followed

later by developing the theoretical bases together with W. S. Franklin [86].

The expression gives the analytical form of the reverberation time obtained by

Sabine:

0.161V

S ¯α

[s] (2.10)

where V is the volume of the room and S ¯α represents the amount of absorption

present in the environment, the unit is named [Sabins] in honor of Sabine.

Subsequent models improved the calculation of the reverberation time by con-

sidering the evolution of the energy density, and the sound absorption carried

out by the air [74], the specular reﬂection of each sound wave [187, 271], the

propagation path [148], the triaxial arrangement of the diﬀerent absorption

coeﬃcients [14, 82], among others.

In addition to statistical theory, T

can be obtained from the measurement

of the impulse response. In measurements, the T

is obtained considering

limitations regarding the background noise level and the sound source’s max-

imum sound pressure level. Thus, according to the ISO 3382 standard [127],

the measurement’s dynamic range must present the end of the decay at least

15 dB above the background noise and start 5 dB below the maximum. For

example, the sound pressure level required to measure the T

in a room with

a background noise of 30 dB is 110 dB (30 + 15 + 60 + 5).

Chapter 2. Spatial Sound & Virtual Acoustics 35

Linear behavior is noted by observing the square of the energy h

(t) in the

decay curve plotted in dB (See Figure 2.12). Thus, to reduce the dynamic

range required for measurement, it is possible to estimate the T

through

other limits. The T

is commonly mistaken for double the T30, which is not

true. The T

and T

also correspond to the time the sound pressure level

(SPL) inside the room takes to drop 60 dB but estimated from measurement

restricted to ranges of -5 dB to -25 dB, and -5 dB to -35 dB, respectively.

Therefore, the linear energy decay produces the relation T

= T

Figure 2.12: Normalized Room Impulse Response: example from a real room in

the time domain (left), and in the time domain in dB (right).

The T

is obtained as the decay rate by the linear least-squares regression of

the measured decay curve, also called the Schroeder curve, in the range -5 dB

to -25 dB. In comparison, the T

is obtained when the curves’ adjustment is

carried out in the range between 5 dB and -35 dB [127].

2.3.3.3 Clarity and Deﬁnition

The clarity and deﬁnition parameters express a balance between the energy

that arrives earlier and later in the impulsive response, which is related to

Chapter 2. Spatial Sound & Virtual Acoustics 36

human beings’ particular ability to distinguish sounds in sequence [44, 45, 57,

247, 254]. With the ﬁrst reﬂections arriving within the limits of 50 or 80

milliseconds, the tendency is to be integrated by the auditory system into the

direct sound. Thus, if the ﬁrst reﬂections contain relatively greater energy

than the reverberating tail, the sound will be experienced as ampliﬁed. On

the other hand, if the reverberating tail has more energy and is long enough,

it will be perceived and mask the next direct sound. The limits of 50 and 80

milliseconds are deﬁned in the literature as appropriate in optimizing speech

and music, respectively [245, 247].

The Clarity deﬁned in the ISO 3382 standard measures the ratio between the

energy in the ﬁrst reﬂections and the energy in the rest of the impulse response.

Clarity’s positive values, which are given in dB, mean more energy in the ﬁrst

reﬂections. Negative values indicate more energy in the reverberating tail. A

null value indicates the balance between the parts of the impulse response.

The ”Clarity” is given by:

= 10 log

80ms

(t)dt

∞

80ms

(t)dt

(2.11)

and

= 10 log

50ms

(t)dt

∞

50ms

(t)dt

(2.12)

The ”Deﬁnition” parameter, in turn, is presented on a linear scale and com-

putes the ratio between the energy contained in the ﬁrst reﬂections by the

total energy of the impulse response. Values greater than 0.5 indicate that

most of the impulse response’s energy is contained in the ﬁrst reﬂections. The

”Deﬁnition” is given by:

Chapter 2. Spatial Sound & Virtual Acoustics 37

80ms

(t)dt

∞

(t)dt

(2.13)

and

50ms

(t)dt

∞

(t)dt

(2.14)

2.3.3.4 Center Time

The central time is a parameter analogous to the previous ones, measuring the

balance between the energy contained in the early reﬂections and the rever-

berating tail’s energy. However, the central time is particularly interesting in

pointing out what can be seen as the center of gravity of the squared impulse

response. Moreover, the central time does not previously delimit the transition

barrier between ﬁrst reﬂections and a reverberating tail. Thus, the deﬁnition

of the central time for an impulse response is given by:

∞

(t)dt

∞

(t)dt

(2.15)

2.3.3.5 Parameters related to spatiality

The relation to the human auditory spatiality sensation and the objective

parameters derived from measurements are studied in detail in the litera-

ture [20, 21, 88]. They observe how the sound energy distribution is arranged

from the directions and timing aspect. The principal sensations and their

Chapter 2. Spatial Sound & Virtual Acoustics 38

related parameters are presented for better understanding.

Apparent Source Width The Apparent Source Width (ASW) is related

to the impression of the sound source’s size or how the source is distributed in

the space.

An objective metric associated with ASW is the Lateral Energy Fraction

(LEF). The Equation 2.16 gives the LEF:

LEF =

80ms

5ms

(t)dt

80ms

(t)dt

(2.16)

Where h(t) is the impulse response measured with a microphone that has an

omnidirectional sensitivity pattern and h

(t) is the impulse response measured

with a microphone that has bidirectional sensitivity (pressure gradient) at the

same position as the omnidirectional.

Thus, this objective parameter represents the ratio between the lateral energy

that reaches the receptor between 5 and 80 milliseconds (i.e., the energy con-

tained in the early reﬂections, excluding the direct sound) and the total energy

arriving from all directions between 0 and 80 milliseconds [21]. As low and

mid frequencies make the dominant contributions to the LEF, this parame-

ter is usually represented by the arithmetic mean of the octave bands’ values

obtained between 125 Hz and 1000 Hz [45, 254].

Listener Envelopment The Listener Envelopment (LEV) is related to the

impression of being immersed in the room’s reverberant ﬁeld.

From Bradley and Soulodre’s experiments [44] with test participants inside

an anechoic room, the sense of involvement was assessed with loudspeakers.

The authors ﬁnd the LEV associated with the ratio between the lateral energy

Chapter 2. Spatial Sound & Virtual Acoustics 39

and the total energy reaching the receptor. The lateral energy is contained in

the impulse response measured with a bidirectional microphone after the ﬁrst

80 milliseconds. The total energy is deﬁned as the impulse response measured

with an omnidirectional microphone, in free ﬁeld condition, and 10 meters away

from the sound source utilizing the same sound source at the same power. The

ratio is called ”Lateral Strength” (LG) and is given by:

LG =

∞

80ms

(t)dt

∞

(t)dt

(2.17)

Interaural Cross-Correlation Coeﬃcient In his work, Keet [133] pro-

posed an auditory-cognitive process relating the spatial impression to compar-

ing the signals received by both ears.

The cross-correlation function measures the degree of similarity of the signals.

Therefore, the Inter-Aural Cross-Correlation Coeﬃcient (IACC) was incorpo-

rated as a third parameter related to the spatial impression. The IACC is de-

ﬁned as the absolute maximum value of the ratio between the cross-correlation

function of the impulse responses collected from the left ear (h

(t)) and the

right ear (h

(t)) by the total energy contained in each of them.

IACC = max



(t)h

(t + τ)dt



(t)dt



(t)dt





(2.18)

where

(t)dt and

(t)dt are the energy between the instant t1 and

the instant t2 in the impulse response from the left and right ears; the expres-

sion

(t)h

(t + τ)dt is the cross correlation function between the impulse

response; τ is given between 0 and 1 ms.

Chapter 2. Spatial Sound & Virtual Acoustics 40

2.3.4 Loudspeaker-based Virtualization in Auditory Re-

Virtualization of sounds through auralization of simulated environments have

been used in architectural design to preview the sound behavior in rooms when

changing space design or even to preview a completely new space that is not

built yet before the building process [293]. As the room acoustics simulation

and the auralization process evolves as a sound equivalent to the visual preview

rendered in 3D models, it has found applications in research also outside the

architectural ﬁeld [282, 320]. Lately, the virtualization of sound sources has

been applied to extend the ecological validity of sound scenarios in auditory

research [161].

Research that utilizes binaural virtualization with headphones are common in

auditory research literature [4, 38, 142, 248, 305]. A series of advantages may

include, but are not limited to, the individual control of the stimuli reproduced

in each ear, the smaller setup, and easier calibration [251]. Although binaural

reproduction is a suitable method for some research questions, others may

require a more complex test environment, especially research encompassing

hearing aids.

In that regard, the use of loudspeakers can be associated with single loud-

speaker presentations where a loudspeaker reproduce a single sound source

positioned in space (e.g., [176, 230, 268, 321]) or virtualization methods that

manage auralized ﬁles to be perceived as single sources or complex environ-

ments [89, 177, 295].

For the virtualization of sound sources, the loudspeaker number depends on the

selected method of encoding and decoding the spatial information [293]. For

example, a quadrophonic loudspeaker arrangement was found to be suﬃcient

Chapter 2. Spatial Sound & Virtual Acoustics 41

to reproduce a diﬀuse sound ﬁeld to a perceptual spatial impression when

constraining listener movements [117]. However, utilizing Directional audio

coding, Laitinen and Pullkki [150] found that to have an adequate reproduction

of diﬀuse sound, it would be necessary from 12 to 20 loudspeakers.

VBAP and HOA techniques were evaluated with diﬀerent numbers of loud-

speakers in simulations by Grimm et al. [97]. Perceptual localization error

(PLE) was computed for the arrays utilizing these techniques. Eight loud-

speakers were estimated to be suﬃcient in terms of sound source localization.

In the same work, Grimm textitet al. showed that the eﬀects of virtualiza-

tion with VBAP and HOA on hearing-aid beam patterns are present with less

than 18 loudspeakers in a bandwidth of 4 kHz (spatial aliasing higher than

5.7 dB criterion). However, the spectral distances, a weighted sum of the ab-

solute diﬀerences in ripple and spectral slope between virtual and reproduced

sound sources, were all very low, indicating high naturalness when compared

to subjective data from Moore and Tan [191].

Aguirre [1] evaluated VBAP and its variation Vector-Based Intensity Panning

(VBIP) in terms of spatial accuracy with 30 normal-hearing participants within

an array of eight loudspeakers. There was no signiﬁcant diﬀerence among

stimuli (speech, intermittent white noise, and continuous white noise) on both

techniques. It was found that an average PLE around 4

◦

, consistent with the

values simulated by Grimm et al. [97].

Evaluating SNR beneﬁts on HA beamformer algorithms within a spherical

array with 41 loudspeakers, Oreinos and Buchholz [212] found similar results

between the real environment and the auralized one. Reproduction errors in

HOA reproduction to hearing aids were studied in [213]. The reverberation

was found to reduce the time-averaged errors introduced by HOA, implying

that the frequency limit of usable renderings with HOA can be extended in

Chapter 2. Spatial Sound & Virtual Acoustics 42

those environments.

Loudspeaker-based virtualization have been used in a hearing research context

evaluating normal hearing, hearing impaired and hearing aid users through

diﬀerent methods [6–8, 30, 31, 55, 61, 80, 93, 102, 136, 168, 174, 188, 220, 303,

322]. Furthermore, some studies explored the ecological validity of the tech-

niques with subjective responses based on psycho-linguistic measure comparing

in-situ and those virtualized in laboratory [66, 103, 286].

The process of virtualizing sound sources using loudspeakers is complex [282]

and requires a thorough understanding of physical acoustics, psychoacoustics,

signal processing fundamentals, and proper calibration of software and hard-

ware [126, 165, 293]. As a result, research centers have developed systems to

establish reliable procedures for virtualizing sound sources for auditory testing.

Examples of such systems include the transaural CTC system developed by

Asp¨ock et al. [17], the system with a spherical array of 42 loudspeakers capable

of rendering scenarios using HOA up to ﬁfth order and VBAP presented by

Parsehian et al. [215], and the Loudspeaker Based Room Auralization (LoRa)

system developed by Favrot [79], which is capable of rendering auditory scenes

using pure HOA and a hybrid version with Nearest Speaker (NSP) and HOA.

In addition, Grimm [100, 101] introduced the Toolbox for Acoustic Scene Cre-

ation and Rendering (TASCAR), which is capable of rendering perceptually

plausible scenes in real-time using VBAP and HOA 2D implementation. A

recent study by Hamdan and Fletcher [107] proposed a method using only two

loudspeakers in 2022 based on the transaural method with cross-talk cancella-

tion. While this list is not exhaustive, these studies provide recommendations

and guidelines for the ﬁeld and highlight the importance of implementing re-

liable systems and verifying their sound ﬁelds objectively and subjectively to

increase the ecological validity of auditory research and hearing aid develop-

ment.

Chapter 2. Spatial Sound & Virtual Acoustics 43

2.3.4.1 Hybrid Methods

Hybrid methods that combine the reproduction of direct sound and reverber-

ation are not new, having been developed since at least the 1980s with the

Ambiophonics group [42, 95]. They proposed the Ambiophonics method to re-

produce concerts to one or two home listeners as if they were in the hall where

the recording was performed. This method combined crosstalk canceled stereo-

dipole and convolved signals with the IR from the recorded spaces [76, 94]. The

system aims to enhance the reproduction of recordings from existing systems

(e.g., stereo and 5.1). The group also developed a new recording methodology

called Ambiophone. This method is a microphone arrangement composed of

two head-spaced omnidirectional microphones covered by a baﬄe in the rear

to favor room reﬂections from frontal directions.

In 2010, the Loudspeaker based Room Auralization (LoRa) method developed

by Favrot [79] applied the hybrid concept using HOA and the nearest speaker

(NSP) for the direct sound and early reﬂections. The method uses the enve-

lope from simulated rooms to reduce the computational cost by multiplying it

with uncorrelated noise. The scheme was originally conceptualized for a large

spherical 69 loudspeaker array. Figure 2.13 depicts its system schematic.

Figure 2.13: LoRa implementation processing diagram. The multichannel RIR is

derived in eight frequency bands and for each part of the input RIR (Figure from

Favrot [79]).

Chapter 2. Spatial Sound & Virtual Acoustics 44

Pelzer et al. [221] presented a comparison between transaural or cross-talk

cancellation (CTC), VBAP, and 4th-order Ambisonics among two new hybrid

proposals: (1) direct sound and early reﬂections through CTC and late re-

ﬂections with 4th-order Ambisonics and (2) direct sound and early reﬂections

through VBAP and late reﬂections with 4th-order Ambisonics. The hybrid

methods were implemented in a single case without generalization to diﬀerent

simulations. These methods were tested within a 24 loudspeaker array with no

statistically signiﬁcant change in human localization performance by any of the

methods. Pausch et al. [217] presented a method designed for investigations

with subjects with hearing loss. The method mixes binaural techniques to

process components in complex simulated environments and CTC to present

them over loudspeakers. At the same time, the head position can be tracked,

allowing user interaction.

In 2017 Pulkki et al. [243] presented the ﬁrst-order directional audio coding

(DirAC) method is a technique for reproducing spatial sound over a standard

stereo audio system. It is based on using ﬁrst-order ambisonic channels, which

encode the sound pressure and particle velocity at a listener’s location to repre-

sent the sound ﬁeld. These channels are transformed into a stereo audio signal

using a frequency-dependent matrix, which preserves the spatial cues that are

important for localizing sound sources. The method implies the direction of ar-

rival of the sound source to be able to virtualize it through amplitude panning.

It uses real-world recordings.

The DirAC method is eﬀective for various types of audio content, including

music, speech, and sound eﬀects. It can potentially improve the spatial realism

of audio experiences over traditional stereo systems and has applications in

myriad ﬁelds, including entertainment, gaming, and virtual reality.

Table 2.1 presents an overview of the listed methods and the techniques in-

Chapter 2. Spatial Sound & Virtual Acoustics 45

volved, their purpose, and their parameters.

Table 2.1: Non-exhaustive overview list of hybrid auralization methods proposed

in the literature. The A-B order of the techniques does not represent any order of

signiﬁcance.

Year Method Authors

Technique

Proposed

Loudspeaker

Number

Proposed to

1986 Ambiophonics Farina et al. [76]

Crosstalk

Cancelation

Binaural 2 Music Reproduction

2005 DirAC Pulkki et al. [243]

Ambisonics

VBAP 2+ multiple applications

2010 LoRa Favrot [79] HOA NSP 64

2014 - Pelzer et al. [221]

Crosstalk

Cancelation

HOA 24

2014 - Pelzer et al. [221] VBAP HOA 24

2018

Extended Binaural

Real-Time

Pausch et al. [217] Binaural CTC 2

Hearing Loss

Investigations

2.3.4.2 Sound Source Localization

A comparison among VBAP and Ambisonics conducted by Frank [84] demon-

strated a median deviation in experimental results from the ideal localization

curve of 2.35º± 2.93º on VBAP and 1.05º± 4.07º to third order Ambisonics

using max-r

decoder. The setup was placed in a typical non-anechoic studio

and a regular array of 8 loudspeakers, listening in a 2.5m radius circle at the

central position. The subjective results from 14 participants were listening to

pink noise. These experimental results were compared to a localization model

(Lindemann [157]) based on ITD and ILD from impulse responses. The re-

sults showed a deviation close to the standard deviation in subjective listening,

2.35º on VBAP and 3.37º to third order Ambisonics using max-r

. Oﬀ-center

measurements were pointed out as necessary for future investigation by the

author.

Ambisonics in ﬁrst, third, and ﬁfth order was examined in another study by

Frank and Zotter [85], with 15 normal hearing, 12 loudspeakers setup, and

listening to pink noise with interval attenuation. This study investigated the

eﬀect of the position (centered and oﬀ-center) and the order. The results

showed, to the ﬁrst order rendered, a localization error of around 5º to the

Chapter 2. Spatial Sound & Virtual Acoustics 46

centered listener and 30º for the oﬀ-center position.

Also, Ambisonics in the ﬁrst order with four loudspeakers and the third order

with eight loudspeakers was investigated by Stitt et al., [283]. This study was

conducted in a non-reverberant environment to verify the oﬀ-center position

and the Ambisonics order. The setup was a circular array with 2.2 meters

of radius and an RT of 0.095 s. Eighteen test participants listened to white

noise bursts of 0.2 s. At this acoustically dry condition, the centered ﬁrst-order

median absolute error was around 10º, while in the oﬀ-center positions tested

was close to 30º. As expected, the error was lower in the third order achieving

a median of absolute error around 8º in the center and 11º oﬀ-center.

A study by Laurent et al., [275] investigated the eﬀect of 3D audio repro-

duction artifacts on hearing devices assessing ITD, ILD, and DI on HOA

(third and ﬁfth orders), VBAP, distance-based amplitude panning (DBAP),

and multiple-direction amplitude panning (MDAP). The study was conducted

in a non-anechoic room with 32 loudspeakers in a spherical conﬁguration. The

loudspeaker distance from the center was 1.5 m, except for the four loudspeak-

ers at the top, which were distant only 98 cm. This study investigated centered

and oﬀ-center positions (10 and 20 cm). The results presented an expected

Ambisonics limitation of reproducing ITD because of the spatial aliasing at

high frequencies, accordingly to the authors. In addition, they investigated

MVDR monoaural beamformer, which did not reproduce the correct ITD, es-

pecially oﬀ-center. At the centered position, only DBAP could not correct

reproduced ITD. Ambisonics ITDs deteriorate more than VBAP on oﬀ-center

positions. ILD errors in virtualized sound sources can make the system unre-

liable for testing HI with processing based on ILDs. In the experiment, the

ILDs were less aﬀected by beamforming processing in VBAP, and Ambisonics

beneﬁted from the max

decoding that maximizes the energy vector. How-

ever, the authors expect a better ILD representation from VBAP as HOA has

Chapter 2. Spatial Sound & Virtual Acoustics 47

an aliasing frequency limitation.

Hamdan and Fletcher [107] present the development of a compact two-loudspeaker

virtual sound reproduction system for clinical testing of spatial hearing with

hearing-assistive devices. The system is based on the transaural method with

cross-talk cancellation and is suitable for use in small, reverberant spaces, such

as clinics and small research labs. The authors evaluated the system’s perfor-

mance regarding the accuracy of sound pressure reproduction in the frontal

hemisphere. They found that it could produce virtual sound ﬁelds up to 8kHz.

They suggest that tracking the listener’s position could improve the system’s

performance. Overall, the authors believe this system is a promising tool for

the clinical testing of spatial hearing with hearing-assistive devices.

Finally, a study by Bates et al., [22], evaluated second-order Ambisonics and

VBAP localization errors in subjective listening tests and ITD and IAFC com-

parisons. They presented the stimuli to a simultaneous set of nine listeners in

diﬀerent positions inside a concert room (around 1 s of RT). With 16 loud-

speakers, they selected 1 second of speech (male and female), white noise, and

music. The results indicate that VBAP and Ambisonics techniques cannot

consistently create spatially accurate virtual sources for a distributed audience

in a reverberant environment. The oﬀ-center positions are compromised by

technique and stimulus. Depending on the stimuli, centered positions resulted

in localization errors between 10º and 20º degrees. In the spatial distribution

inside the ring, a bias from the target image position and towards the nearer

contributing loudspeaker is more present in the Ambisonics than the VBAP.

The authors mentioned that the room acoustics could also impact localization

accuracy.

The number of variables across these previous studies and their contributions

is massive, e.g., objective measures, technique variations, number of loud-

Chapter 2. Spatial Sound & Virtual Acoustics 48

speakers, loudspeaker distance, number of simultaneous listeners, reverbera-

tion time, and form of the array.

Table 2.2 presents an overview of methods and estimated or measured local-

ization error.

Chapter 2. Spatial Sound & Virtual Acoustics 49

Table 2.2: Overview of Localization Error Estimates or Measurements from Loudspeaker-Based Virtualization Systems Using Various Auralization

Methods.

Method

Error at

center position

Error at

oﬀ-center position

LS number

Present Study

Iceberg

VBAP/Ambisonics

30º (Max Estimated)

7º (Average Estimated)

30º (Max Estimated)

7º (Average Estimated)

Frank [84] VBAP 2.35º (Average) N/A 8

Frank [84] HOA (3rd order) 3.37º (Average) N/A 8

Zotter [85] Ambisonics 5º (Median) 30º (Median) 12

Zotter [85] HOA (3rd order) 2º (Median) 15º (Median) 12

Zotter [85] HOA (5thorder) 1º (Median) 10º (Median) 12

Stitt et al., [283] Ambisonics 10º (Median) 30º (Median) 8

Stitt et al., [283] HOA (3rd order) 8º (Median) 11º (Median) 8

Bates et al., [22] Ambisonics Second Order 10º (mean) 20º (mean) 16

Bates et al., [22] VBAP 10º (mean) 20º (mean) 16

Grimm et al. [97] HOA (3rd order) 2º (Estimated) 6º (Estimated) 8

Grimm et al. [97] VBAP 4º (Estimated) 6º (Estimated) 8

Aguirre [1] VBAP 4º (Median) N/A 8

Hamdan and Fletcher [107] CTC 2º (Max Head Displacement) N/A 2

Huisman et al. [125] Ambisonics 30º (Median) N/A 4

Huisman et al. [125] HOA (3rd order) ≈ 15º (Median) N/A 8

Huisman et al. [125] HOA (5th order) ≈ 8º (Median) N/A 12

Huisman et al. [125] HOA (11th order) ≈ 5º (Median) N/A 24

Chapter 2. Listening Eﬀort Assessment 50

2.4 Listening Eﬀort Assessment

The regular task of following a conversation, listening to a person’s speech,

or interacting with someone in a conversation may require additional eﬀort in

an unfavorable or challenging sound environment [227]. The listening eﬀort is

deﬁned as ”the deliberate allocation of mental resources to overcome obstacles

in goal pursuit when carrying out a [listening] task” [224]. Studying aspects

of the listening eﬀort related to diﬀerent acoustic situations through reliable

methods can lead to the development of solutions to reduce it, improving the

quality of life [304]. However, there is no consensus in the literature on the

best method to measure listening eﬀort.

Attempts to measure how much energy a person takes in a speciﬁc acoustic sit-

uation may rely on diﬀerent paradigms. There are objective measurements of

physiological parameters in literature associated with changes in eﬀort, such

as pupil dilation [151, 209, 211, 301, 302, 319], responses to brainstem fre-

quency (FFRs) and cortical electroencephalogram (EEG) activity from event-

related potentials [28, 33], or alpha band oscillations [186, 223]. In addition,

the behavioral perspective studies changes in response time in single [204] or

dual-task paradigm tests, also assuming that they are related to changes in

cognitive load in auditory tests [87, 228] [225]. In turn, subjective assess-

ments of listening eﬀort are performed through questionnaires [323] or eﬀort

scales [147, 149, 249, 260] and their results generally agree with performance

metrics [192].

Although subjective measurements are intuitive and valid, they tend to be less

accepted as an indication of the amount of listening eﬀort because of diﬀerences

between objective and subjective outcomes [151, 225]. For instance, Zekveld

and Kramer [318] present evidence of disagreement between the physiological

and the subjective measure where the young normal-hearing participants at-

Chapter 2. Listening Eﬀort Assessment 51

tributed high subjective eﬀort to the most challenging conditions despite their

smaller pupil dilation. The authors assumed that the methodological aspects

and the participant’s tendency to drop out were also related to pupil dilation at

low levels of intelligibility. In a study on syntactic complexity and noise level in

the auditory eﬀort, Wendt et al. [300] evaluated it through self-rated eﬀort and

pupil dilation. They found both background noise and syntactic complexity

reﬂected in their measurements. However, at high levels of intelligibility, the

methods show diﬀerent results. According to the authors, the explanation is

that each measure represents a diﬀerent aspect of the eﬀort. In its turn, Picou

et al. [226]; and Picou and Ricketts [229] found subjective ratings of listening

eﬀort were correlated with performance instead of the listening eﬀort utilizing

the response time as a behavioral measure in a dual-task. Interestingly though,

in this study, a question about control was correlated to their response time

results. The varied outcomes from subjective and objective paradigms pro-

posed to achieve a proxy to listening eﬀort can indicate that these methods

are quantifying separated aspects of a complex global process [12, 224].

Another explanation suggests a bias in the subjective method due to the heuris-

tic strategies adopted by the participants to minimize the eﬀort [192]. The

mentioned strategy would consist of replacing the question about the amount

of eﬀort spent with a more straightforward question related to how they per-

formed in the task. Concomitantly, studies based on objective measurement

paradigms also have divergent results. For example, even among physiological

measures sensitive to the spectral content of stimuli, such as pupil dilation and

alpha power, they are not always related, and can be sensitive to diﬀerent as-

pects of listening eﬀort [186]. Even within the same paradigm, a diﬀerent task

may indicate that diﬀerent aspects are being observed. For example, Brown

and Strand [53] analyzed the role of the working memory as a weighting factor

on listening eﬀort. Although increasing background noise indeed increases lis-

tening eﬀort measured by the dual-task paradigm, the memory load was not

Chapter 2. Listening Eﬀort Assessment 52

aﬀected. They also suggested that the working memory and listening eﬀort

are related in the recall-based single-task, unlike in the dual-task. In Lau et

al. [151] signiﬁcant diﬀerences between sentence recognition and word recog-

nition were found on pupil dilation measurements and in subjective ratings,

although with no correlation between objective and subjective measures.

The demand for mental resources can also be aﬀected by personal factors, such

as fatigue and motivation [224]. At the same time, several physical-acoustical

artifacts can degrade a sound, creating or leading to diﬃculties in everyday

communication (increasing listening eﬀort), especially in social situations. The

masking noise, the spectral content of the noise, the Signal-to-Noise Ratio

(SNR), and the environment reverberation are examples of artifacts capable

of smearing the temporal envelope cues [163].

Also, speech intelligibility was assessed in a virtual environment that consists in

a large spherical array of 64 loudspeakers reproducing Mixed Order Ambisonics

(MOA) [6] presented comparable results of Speech Reception Threshold (SRT)

compared to real room in a co-located situation of masker and target. With

spatial separation of 30 degrees the virtual environment led to an SRT beneﬁt

of 3 dB, it was argued that beneﬁt was not present in more reverberant or

complex scenes suggesting the masking eﬀect of more challenge scenes.

SRTs for normal hearing and hearing-impaired using hearing aids were also in-

vestigated by [31]. A complex scenario (reverberant cafeteria) and an anechoic

situation were evaluated in a spherical array of 41 loudspeakers. The virtu-

alization was provided convolving the direct sound and the early reﬂections

parts of the RIR with the anechoic sentence and presenting the sound through

the Nearest Speaker (NSP) and the late reﬂections part of the RIR are created

through the directional envelope of each loudspeaker with uncorrelated noise.

Chapter 2. Concluding Remarks 53

The reviewed studies were conducted in laboratories mainly taking advantage

of spatial sound and virtual acoustics via loudspeaker or headphones repro-

duction. Thus, the complex nature of human auditory phenomena and the

importance of reproducibility in hearing research highlight the need for inno-

vative tools such as spatial sound [134]. Virtualized sounds allows for realistic

and controllable sound environments, enabling control over selected param-

eters and consistent reproduction of experiments [61, 161, 282, 293]. This

technology can help hearing investigations become more true-to-life and reli-

able [134, 161, 251]. For example, it can be used to study listening eﬀort and

speech intelligibility using virtual sound sources to create ecologically valid

and controlled environments [7, 177]. It also can enable the integration of vir-

tual sound scenarios with ecological tasks involving multiple people, providing

an ecologically valid assessment of the performance of hearing solutions more

accessible than large-ﬁeld studies (e.g., in Bates et al., [22]).

Additionally, spatial audio enables the accessible investigation of spatial sep-

aration’s eﬀects on binaural cues considering diﬀerent environments, the role

of binaural hearing in spatial perception, and new hearing aid hardware and

algorithms [61, 97, 213]. Overall, spatial sound & virtual acoustics in hearing

research oﬀers numerous beneﬁts and represents a valuable tool for advancing

our understanding of hearing and developing eﬀective hearing solutions.

2.5 Concluding Remarks

The literature review suggests a contrast between localization and immersion in

auralization methods that virtualize sound using a low number of loudspeakers.

Thus, there is a need for a method that can achieve useful performance on

both localization and immersion with a small number of loudspeakers and

that is reliable in rendering sound for listeners in the presence of another

Chapter 2. Concluding Remarks 54

listener within the virtualized sound ﬁeld. Previous methods, including hybrid

approaches, have been developed using a larger number of loudspeakers and

diﬀerent techniques for balancing energy. A recent study proposed a method

using only two loudspeakers in 2022. However, it implemented a diﬀerent

auralization method and had its limitations. The proposed method in this

study is innovative, using a room acoustic parameter called center time to

calculate the energy balance of room impulse responses and combining it with

two known auralization methods.

Chapter 3

Binaural cue distortions in

virtualized Ambisonics and

VBAP

3.1 Introduction

In acoustics, the complex communication scenarios can involve, simultaneous

sound sources, distracting background noise, moving sound sources, sources

without large spatial separation, and low signal to noise ratio. Although peo-

ple with normal hearing can deal with most of these conditions in a relatively

eﬃcient way, people with hearing loss perform poorly [273, 289, 317]. Since so-

cial events are often a real example of complex communication, the interaction

barriers make people avoid this and sometimes ostracizing themselves [16, 63].

That can be a factor in decreasing the quality of life of people with hearing

problems.

In hearing research, innovative signal processing techniques, new devices, more

Chapter 3. Introduction 56

powerful hardware, and updated parameter settings are continuously devel-

oped and evaluated. These technological improvements aspire to resolve com-

munication problems in everyday situations for hearing aid users [227], increas-

ing their socialization and quality of life [119]. Tests as speech recognition in

noise are developed and tailored to evaluate the human auditory response on

everyday acoustics situations better than clinical based in pure tones stim-

ulation [145]. Even though the tasks are moving towards a more realistic

representation, they still need to improve the ecological validity [134].

Auralization methods are designed to create ﬁles meant to be reproduced to a

speciﬁc listener or a group of listeners; these ﬁles contain particular character-

istics that try to mimic a recorded or digitally created sound scene according to

the method. The mathematical formulations that produce these characteristics

for the psychoacoustically based methods focus on delivering accurate binaural

cues. The listener position, physical obstacles as the listener movement will

impact diﬀerently on distinct methods and cues.

A VSE is an auralized sound ﬁeld that can contain realistic elements. Cur-

rently, it is possible to create VSE employing loudspeaker arrays or headphones

for the listener, such as high background noise, high reverberation, and con-

comitant sound events from diﬀerent directions [61, 79, 294]. Furthermore,

through a VSE, it is also possible to enable a participant to wear, for example,

a hearing aid during the test. Thus, the researcher can maintain control of

the stimuli, the incidence direction, signal-to-noise ratio (SNR), among other

settings, while examining the hearing device performance in a more ecological

situation [98, 161, 269].

Although novel technologies emerge and contribute to emulating sound sources

and even entire complex sound scenes with humans’ social interaction [267],

these opportunities are often overlooked in auditory evaluations. Typically,

Chapter 3. Introduction 57

tests are performed by observing only one individual within the laboratory [81,

89, 104, 152, 169, 175]. Furthermore, the systems are designed to acquire

responses from a single individual at a time [41, 79, 102, 118, 195, 218–220, 259].

A reasonable explanation for this is the low cost and complexity of auralization

through headphones. More complex techniques, like Wave Field Synthesis, do

not limit the listener to a restricted spot [207], reproducing a complete sound

ﬁeld, although at the cost of a large number of sound sources in a speciﬁcally

treated room.

Social situations can have eﬀect on people’s listening eﬀort [230, 234] and their

motivation to listen [181, 224]. In this context, social interactions have been

simulated through avatars or audiovisual recordings in virtual environments,

gaining space in auditory research [116, 160, 161, 272, 298]. Although it can be

considered a signiﬁcant asset, it also focuses on a single individual’s responses

to simulated social stimuli.

The scenario creates a ground for this study to investigate controlled acousti-

cal changes on the VSE. This study assesses two main situations within a ring

of loudspeakers virtualizing sound sources on Ambisonics and VBAP: (1) the

displacement of the listener from the center (sweet spot), and (2) the eﬀect

including a second simultaneous listener inside the ring. These topics can help

understand the perception of sound in these speciﬁc virtualization methods,

increasing the fundamental scientiﬁc basis for future hearing research appli-

cations. The changes to the sound ﬁeld were observed in three major spatial

cues: ITD, ILD, and IACC. That was explored by changing the listener’s posi-

tion and including a second listener inside the ring of loudspeakers to measure

BRIRs.

These metrics can describe the spatial perception of an auralized sound sig-

nal [47, 48], being ITD and ILD responsible by localization and IACC perceived

Chapter 3. Methods 58

spaciousness and the listener envelopment [44]. Therefore, these measurements

can indicate the possibility of a simultaneous second participant in any hear-

ing test with virtualized spatially distributed sound sources. Two diﬀerent

auralization techniques were used to virtualized sound sources, vector-based

amplitude panning (VBAP) [241] and Ambisonics [91]. Both techniques rely

on the same receptor-dependent psychoacoustic paradigm to provide an au-

ditory sense of immersion for those with normal hearing [161, 180]. These

techniques aim delivering the correct binaural cues to a point or area to create

a realistic spatial sound impression, albeit through diﬀerent mathematical for-

mulations. The work investigates if the techniques can provide an appropriate

spatial impression for young normal-hearing listeners.

Hypothesis The main research question is how auralized scenarios with

VBAP and Ambisonics are aﬀected when displaced from the center and with

another listener inside the ring. The hypothesis is that localization cues can be

better provided by VBAP, especially in oﬀ-center positions. In contrast, Am-

bisonics can provide a better sense of immersiveness. Also, the second listener

would impact Ambisonics more than VBAP virtualized sound sources.

3.2 Methods

The experiment was conducted in two diﬀerent locations; The ﬁrst one is a

sound treated test room at the Hearing Sciences - Scottish Section in Glasgow

(See Figure 3.1 ), the second is an anechoic test room at Eriksholm Research

Centre (See Figure 3.2). This section presents the rooms’ acoustic characteri-

zations and the methods used in this experiment.

Chapter 3. Methods 59

Figure 3.1: Hearing Sciences - Scottish Section Test Room.

Figure 3.2: Eriksholm Test Room.

3.2.1 Setups and system characterization

The experiment conducted in Glasgow was in a large sound-proof audiometric

booth (4.3 × 4.7 × 2.9 m; IAC Acoustics). An azimuthal circular array conﬁg-

uration of 24 loudspeakers (3.5-m diameter; 15

◦

of separation; Tannoy VX6)

was used. The ceiling and walls were covered with 100-mm deep acoustic foam

wedges to reduce reﬂections; the ﬂoor was carpeted with a foam underlay. The

AD/DA audio interface that was used was a Ferroﬁsh Model A32. The loud-

speakers received signals that were ampliﬁed by ART SLA4 ampliﬁers. The

reference microphone used to characterize the Glasgow Test Room was a 1/2”

G.R.A.S 40AD pressure-ﬁeld microphone set with e GRAS 26CA preampliﬁer.

It was oriented 90 degrees vertically from the sound source. At Eriksholm,

an equivalent setup was ﬁtted. This time in a full anechoic room from IAC

Chapter 3. Methods 60

Acoustics. The room’s outer dimensions (6.7 × 5.8 × 4.9 m; ) and inner di-

mensions, from the tip of the foam edges (4.3 × 3.4 × 2.7 m). An azimuthal

circular array conﬁguration of 24 active loudspeakers (16 Genelec 8030A and

8 Genelec 8030C; 2.4-m diameter; 15

◦

of separation) was used. The AD/DA

was a MOTU PCI-e 424 combined with a ﬁrewire 24-channel audio extension.

The reference microphone used to characterize the Eriksholm test room was a

1/2” B&K 4192 pressure-ﬁeld and a preampliﬁer type 2669, supplied by power

module 5935. It was oriented 90 degrees vertically from the sound source.

The signal acquisition and processing were entirely through Matlab 2020a soft-

ware using the ITA-Toolbox v.9 [29].

The technical setup was equivalent in both rooms, a B&K head and torso sim-

ulator (HATS) model 4128-C mannequin for measurements, and a Knowles

Electronics Mannequin for Acoustic Research (KEMAR) was used as a phys-

ical obstacle. Although technically, both devices are head and torso simulators,

in this thesis, HATS will refer to the B&K 4128-C for simplicity. The sampling

rate of the recordings was ﬁxed at 48 kHz, resulting in an uncertainty of ±20

µs, therefore not compromising the ﬁnal analysis.

3.2.1.1 Reverberation time

The reverberation time is one of the most critical objective parameters of a

room [154]. The decay of sound energy to 60 dB below its peak after the

cessation of a sound source characterizes the RT. The parameter is frequency-

dependent; it is associated with speech understanding speech, sound quality,

and the subjective perception of the size of the room. For controlled environ-

ments, the values are fractions of seconds. The T

for both rooms in the third

octave is presented in Figure 3.3.

Chapter 3. Methods 61

Figure 3.3: Reverberation time in third of octave bands up to 16 kHz.

The room’s reverberation time T

was measured using a loudspeaker, arbi-

trarily chosen, and microphone setup as in Section 3.2.1. The measurement

and analysis were performed in Matlab through the ITA-Toolbox software.

3.2.1.2 Early-reﬂections

To ensure that there is no inﬂuence of the environment, Recommendation

ITU-R 1116-3:2015 [126], determines that the magnitude of the ﬁrst reﬂections

should be at least 10 dB below the magnitude of the direct sound ∆SPL ≥

10 dB. The diﬀerences in the SPL that are determined in the environments

of this work met this requirement. Table 3.1 shows the diﬀerence in sound

pressure level between the direct sound and early reﬂections. Higher diﬀerences

in the Erkisholm environment are consistent with its anechoic setup compared

to the sound treated booth in Glasgow, where the ﬂoor provide some energy

to the reﬂections.

Chapter 3. Methods 62

Table 3.1: Sound pressure level diﬀerence between direct sound and early reﬂections

∆ SPL [dB]

Angle Eriksholm Glasgow

0 -20.99 -14.94

15 -23.40 -15.31

30 -22.66 -14.61

45 -21.97 -15.45

60 -20.39 -13.28

75 -21.22 -15.19

90 -17.71 -15.33

105 -21.49 -15.22

120 -17.83 -15.68

135 -20.12 -15.23

150 -19.70 -14.62

165 -19.13 -16.11

180 -24.57 -15.03

195 -23.56 -13.52

210 -22.62 -14.81

225 -21.04 -15.39

240 -22.29 -14.25

255 -23.73 -14.37

270 -20.90 -14.01

285 -24.06 -12.56

300 -19.61 -15.95

315 -17.68 -15.03

330 -21.46 -15.66

345 -23.08 -15.95

3.2.2 Procedure

The experiment studied how the presence of a second listener within a loud-

speaker ring aﬀects the spatial cues of the reproduced sound ﬁeld. The data

were collected through the HATS, and the second listener being simultaneously

inside the virtualized sound area was simulated through another mannequin

(KEMAR), as shown in Figures 3.4 and 3.5.

Using the results for the reverberation time as presented in Section 3.2.1, the

appropriate length of a logarithmic sweep signal was calculated as approxi-

mately four times larger than the higher value of T

(1.49 seconds). Also, a

stop margin of 0.1 seconds was set to ensure the quality of the room impulse

Chapter 3. Methods 63

Figure 3.4: HATS (with motion-tracking crown) and KEMAR inside test room in

Glasgow.

Figure 3.5: HATS and KEMAR inside anechoic test room at Eriksholm.

responses (RIRs) that were obtained [75, 194]. The frequency of the sweep

was from 50 Hz to 20 kHz.

The position of the head has a signiﬁcant eﬀect on the signals that are mea-

sured. To have a reliable assessment of the absolute tri-dimensional position

of the HATS, its position was measured with a the Vicon infra-red tracking

Chapter 3. Methods 64

system with an accuracy of 0.5 mm in Glasgow. At Eriksholm a laser tape

measure was used to ensure the correct positions. The microphones’ height

position in both experiments was set to match the geometrical center of the

loudspeakers enclosure in all measurements. The ﬁrst position measured used

the HATS in the center, without interference from another obstacle inside the

ring, to provide a baseline.

Figure 3.6a illustrates a set of positions to study the inﬂuence of a second

listener inside the ring while keeping the test subject in the center (the sweet

spot). Three diﬀerent positions for the KEMAR (50, 75 and 100 cm of sepa-

ration) were measured with the HATS ﬁxed at the center of the loudspeaker

array. The data collected are from microphones in the HATS ears; the KEMAR

was only a physical obstacle to simulate a listener inside the ring. Figure 3.6b,

illustrates a diﬀerent set of positions, maintaining a minimum separation of

50 cm between the center of the heads, were measured. The purpose of these

positions with the HATS oﬀ-center was to identify the presence of distortions

caused by the decentralization of the subject and the eﬀect of the addition of a

listener within the circle of loudspeakers as a physical obstacle to sound waves.

The positioning was standardized so that the movement along the x-axis to

the left and right directions of the dummies were annotated as negative and

positive, respectively.

3.2.3 Calibration

To calibrate the HATS recordings, the adapter B&K UA-1546 was connected

to the B&K 4231 calibrator. That provided a 97.1 dB SPL signal, which

corresponds to 1.43 Pa, instead of 94 dB without an adapter. The recorded

signal from each ear was used to calibrate the levels of all measurements. The

calibration factor was calculated as:

Chapter 3. Methods 65

(a) Centered position (b) Oﬀ-center position

Figure 3.6: HATS in gray, KEMAR in yellow. a) Measured positions with the

HATS centered and the KEMAR present in the room in diﬀerent positions (three

combinations). b) Measured positions with the HATS in diﬀerent positions and the

KEMAR present in the room in diﬀerent positions (nine combinations).

l,rms

1.43

rms(v

(t)

1kHz

)



VFS



, (3.1a)

r,rms

1.43

rms(v

(t)

1kHz

)



VFS



, (3.1b)

where

l,rms

is the calibration factor for the left ear;

r,rms

is that for the right ear;

(t) is the calibrator signal recorded in the left ear;

(t) is that for the right ear;

The individual loudspeakers’ sound pressure level to the same ﬁle can diﬀer

depending on several factors (e.g., the ampliﬁcation system’s level). To balance

that, a factor was then measured for a GRAS 1/2” pressure-ﬁeld microphone

recording a pistonphone’s calibrated sound signal from 1 kHz. The calibration

factor α

rms

was calculated from the root mean square (RMS) using:

Chapter 3. Methods 66

rms

RMS(v(t)

1kHz

)



VFS



, (3.2)

where

v(t)

1kHz

is the sinusoidal signal at 10 Pa recorded from the calibrator in volts

full scale (VFS). The loudspeaker correction factor is calculated through the

iterative process that starts reproducing a RMS scaled version of a pink noise

signal at 70 dB SPL.

pink noise(t) =



pink noise(t)

rms(pink noise(t))

70−dBperV

20µ



(3.3)

where Γ

is the level factor to the loudspeaker l with initial value = 1;

dBperV = 20 log



rms

20µ



The signal pink noise(t) is played through a loudspeaker l and simultaneously

recorded with the microphone S

(t); the SPL of the recorded signal is calculated

as follows

SPL

[dB] = 20 log

(t)[VFS]α

rms



VFS



20[µPa]

, (3.4)

Ten measurements are sequentially performed, making intervals of 1 second;

the next iteration happens if the SPL obtained exceeds the tolerance of 0.5 [dB]

on any of the measurements. A step of ± 0.1 [VFS] is set to update Γ

in its

next iteration accordingly to the SPL obtained.

Chapter 3. Methods 67

3.2.4 VBAP Auralization

In the ﬁrst measurement, VBAP was the technique used to auralize the ﬁles.

The ﬁrst step in signal processing was recording the 24 (RIRs) one from each

loudspeaker. Knowing the RT of the room, a sweep (50-20000 Hz) was cre-

ated, fulﬁlling the length requirement; in this case, a logarithmic sweep of

1.49 seconds. After that, an inverse ﬁlter (minimum-phased) was created to

compensate for the frequency responses from the diﬀerent loudspeakers. This

signal is then processed through the VBAP technique to the speciﬁed array of

24 loudspeakers. The output is a ﬁle with 24 channels containing the sweep

signal appropriately weighted to the speciﬁc angle. The signal can be processed

through a single channel (when the angle to be played is at the loudspeaker

position) or up to two combined channels when it is a virtual loudspeaker’s

position. Each channel was also convolved with the designed ﬁlter. The ﬁnal

(auralized) signal was used as an excitation in the transfer function where the

receptors were a pair of microphones in the B&K HATS.

3.2.5 Ambisonics Auralization

In the second measurement at the Eriksholm test room, the ﬁles were auralized

with ﬁrst-order Ambisonics in a similarlly to VBAP. To be able to process the

excitation signal, to acquire the impulse responses, some adaptations were

required.

In this case, the Ambisonics auralization process requires an encoded impulse

response that contains the magnitude and the direction of incidence infor-

mation for each instance of time. This RIR can be attained via computer

simulation or recorded with a speciﬁc array of microphones. The ODEON

software version 12.15 was used to simulate the sound behavior in an ane-

Chapter 3. Methods 68

choic environment and encode the impulse responses in Ambisonics ﬁrst-order

format around the listener.

Odeon software is based on a hybrid numeric method [59]. In general, the

Image-Source, a deterministic method, is favored in the region of the ﬁrst

reﬂections up to an order predetermined by the user. Then, reﬂections from

subsequent orders than the predetermined transition order are calculated using

ray tracing, a stochastic method [148, 201]. Therefore, it is possible to simulate

the sound behavior from a 3D model description of the space and details of its

acoustic properties. From that simulation result, any music or sound can be

exported as recorded inside that space from the given positions of source and

receptor [288]. Another option is to export the room impulse response, which

represents the sound behavior of the given source receptor positions. The RIR

can also be exported as BRIR and Ambisonics in ﬁrst and second order in the

version 12 of the Odeon software.

The selected materials used to compose the simulation, and their correspon-

dent absorption coeﬃcients used in the ODEON simulation are listed in the

Appendix E. In total, 72 diﬀerent RIRs (5 degrees separation) were simu-

lated for diﬀerent positions of source-receptor. The simulated source positions

were at the same distance of 1.35 meters from the center as the loudspeak-

ers in the anechoic room. These RIRs were convolved with the appropriate

sweep signal, producing a four-channel ﬁrst-order Ambisonics sweep signal.

These signals were then processed by a decoder to the loudspeakers array’s

speciﬁc positions, generating the auralized 24 channel ﬁles. The inverse ﬁlter

procedure to each loudspeaker was applied as well as the calibration of the

sound pressure level across loudspeakers. The alpha factor was calculated as

rms

rms(v(t)

1kHz

)



VFS



, since the recorded input was from a sound calibrator

type 4231 by B&K delivering 1 [Pa] SPL. The equalized, convolved, decoded,

and ﬁltered sweep signals contain the simulated source-receptor sound distri-

Chapter 3. Results 69

butions in magnitude, time, and space as if recorded inside the simulated room.

In this experiment, the simulated anechoic room has an absorption coeﬃcient

equal to one on all surfaces, simulating the anechoic condition.

The setup in ﬁrst-order Ambisonics was chosen given the possibility of explor-

ing a reduction in the number of loudspeakers in future experiments and the

possibility of generating it through validated software such as Odeon.

3.3 Results

In this study, the performance of the system was evaluated by collecting and

analyzing results based on the positions of a mannequin within the virtual

sound ﬁeld (i.e., center and oﬀ-center) and the conditions under which the

system was tested (i.e., with and without the presence of a second head-and-

torso simulator). The results were presented in terms of angles referenced

counter-clockwise, which allowed for a detailed analysis of the system’s per-

formance under various conditions. Through this analysis, it was possible to

gain a comprehensive understanding of the system’s capabilities and identify

potential areas for improvement.

3.3.1 Analysis

The signals were played and simultaneously recorded; the recorded result car-

ried the auditory spatial eﬀects from auralization and also the physical limita-

tions given by the virtualization setup (e.g., loudspeakers’ frequency response,

and presence of loudspeakers inside the room). As the recorded sweep has

a greater length than the original one, zero-padding was performed. In that

process, zeroes are appended to the end of the time domain signal, obtaining

Chapter 3. Results 70

the equivalent convolution nonetheless [242]. After that, it was possible to

calculate the virtual environment’s impulse response by dividing the recorded

signal by the zero-padded version of the initial sweep, both in the frequency

domain.

For both measurements, the interaural time diﬀerence is calculated by compar-

ing the sound’s arrival time in the impulse response between the two channels

of a binaural room impulse response (BRIR). There are diﬀerent methods for

ITD calculation [132, 314]. In this work, ITDs were estimated as the delay that

corresponds to the maximum of the normalized interaural cross-correlation

function (IACF). According to the ISO-3382-1:2009 [127], the IACF is calcu-

lated as:

IACF

t1,t2

(τ) =

(t)p

(t + τ)dt



(t)dt



(t)dt



(3.5)

where

(t) is the impulse response at the entrance of the left ear canal;

(t) is that for the right canal;

The interaural cross correlation coeﬃcients, IACC [127], are given by:

IACC

t1,t2

= max|IACF(τ)|, for −1ms < τ < 1ms . (3.6)

Similarly, to calculate the interaural level diﬀerence (ILD), a fast Fourier trans-

form (FFT) is applied to the time domain’s impulse responses, the spectrum is

divided into averaged octave bands, and the ratio in dB between the frequency

magnitudes are calculated as the ILD:

Chapter 3. Results 71

ILD(n) = 20 log





(t)





, (3.7)

where:

n is the given frequency band;

(t) is the bandpassed right impulse response;

(t) that to the left channel.

3.3.2 Centered position

In the centered-position conﬁguration, (Figure 3.6a), the listener remains at

the ideal VSE position (center) to focus on the eﬀect of an added listener inside

the loudspeaker ring. This framework can be valuable to auditory research as it

can be used to analyze group responses to interviews, arguments, collaborative

work, social stress or disputes between individuals in listening tasks.

The IACC to the frontal angle (0

◦

) across frequencies is shown in Figure 3.7.

High values indicate that the system delivers the same signal to both ears.

Conversely, the drop in IACC values at high frequencies can indicate that

the Ambisonics may fail to render speciﬁc frequencies aﬀecting the octave

bands analysis. The IACC values measured across all angles for VBAP and

Ambisonics can be found in Figure 3.8. They indicate that Ambisonics tend

to provide less lateralization in lower frequencies (constant and higher IACC

values) and lower but constant values in high frequencies, possibly translating

to blurred sound localization.

Chapter 3. Results 72

Figure 3.7: Interaural cross correlation as a function of frequency in octave bands

- Frontal angle 0º.

Figure 3.8: Interaural cross correlation for averaged octave bands in Ambisonics

and VBAP techniques represented in polar coordinates.

That can happen due to a tilt in positioning the hats or imprecision from the

virtualization system. For example, a high-frequency sound wave at 8 kHz has

a wavelength of approximately 4 cm and 2 cm at 16 kHz, which means that

even a slight tilt can inﬂuence high-frequency IACC. Furthermore, the inverse

FIR ﬁlter applied was not the inverse broadband signal, but the ﬁltered in

Chapter 3. Results 73

third of octave bands. That decision was a signal processing compromise, as a

broadband ﬁlter would only partially compensate for loudspeakers’ geometry

or phase diﬀerences in high frequencies. This point can be further investigated

as a way to improve Ambisonics reproduction.

There is a relative increase of variations with frequency in VBAP results, which

are present to a lesser extent in the Ambisonics IACC results. That reveals

a diﬃculty from Ambisonics to drive a good sense of localization as a high

coherence level indicates the sound coming from front or back [58]. At the

same time, due to the Ambisonics activation of all available loudspeakers to

render the sound in the sweet spot area, the sense of immersion is higher.

3.3.2.1 Centered ITD

The ITD results presented were obtained after a tenth-order low-pass Butter-

worth ﬁlter (LPF) was applied. The ﬁlter’s cutoﬀ frequency was 1,000 Hz to

approximate the low frequency dominance in ITD [38, 124, 197, 242].

Vector Based Amplitude Panning The light blue line in Figure 3.9 shows

the results for the ITD from the initial setup (HATS alone centered). The sys-

tem presented a magnitude peak in response time of approximately 650 µs,

which corresponds to approximately 22 cm for a wave traveling at the ve-

locity of sound propagation in the air. This distance is comparable to the

distance between HATS microphones (19 cm). It is appropriate to note that

the symmetry of HATS is also presented in the HATS alone results (triangles

in Figure 3.9) providing reassurance in the quality of the data collected.

The HATS was kept in the center of the loudspeaker ring for the next set of

measurements. A second listener’s inﬂuence was then simulated by introduc-

Chapter 3. Results 74

Figure 3.9: a) HATS alone at center. b) Light blue line: HATS alone at center.

Black line: HATS centered and KEMAR at 0.5 m to the right. Blue line: HATS

centered and KEMAR at 0.75 m to the right. Red line: HATS centered and KEMAR

at 1 m to the right.

ing a KEMAR and laterally varying its position along the lateral axis (x-axis).

The results are presented in Figure 3.9. The ITD data obtained from this

experiment make it possible to comprehend that the second mannequin (KE-

MAR) has an impact as an obstacle on the interaural time diﬀerence in the

HATS at the center of the loudspeaker ring.

In the closest position of the second listener (50 cm from the center), there

is a reduction of ITD values (angles between 285 and 305 degrees). Thus,

the maximum diﬀerence is 50 us. That eﬀect is related to the insertion of

the physical obstacle represented by the second listener. As the sound wave

diﬀracts, diﬀerent paths to the listener’s ears are imposed, reducing the sound’s

arrival time between ears. Therefore, the eﬀect should be centered at 270

degrees. However, the second listener was not perfectly aligned to the lateral of

the centered listener. That was a limitation of the experiment as the KEMAR

was placed in an ordinary chair, and its bottom is not ﬂat.

Chapter 3. Results 75

Ambisonics The ITD results for the initial setup (HATS alone centered)

virtualized from Ambisonics auralization are presented in Figure 3.11. The

system showed a magnitude peak in response time, roughly 600 µ, 50 µ lower

than the VBAP method. Another characteristic of Ambisonics ITDs is the ﬂat

behavior around the lateral angles, which is generated mainly by the chosen

order of the Ambisonics auralization. In ﬁrst-order, the horizontal directivity

is determined by the to an intersection of three bi-directional (ﬁgure-eight)

sensitivity patterns, circumvented by a omnidirectional one, as illustrated in

Figure 3.10. That can also limit the localization performance when utilizing

ﬁrst-order Ambisonics, even when reproduced through a higher number of

loudspeakers.

Figure 3.10: Horizontal 2D Ambisonics directional sensitivity crop representation.

The red line represents an omnidirectional pattern, the black line represents a bidi-

rectional pattern, y-axis oriented (null points at the sides), and the purple line is

a bidirectional pattern representation x-axis oriented (null points in front and the

back).

The HATS was kept in the center of the loudspeakers ring and simulated a

second listener’s inﬂuence on the sound ﬁeld by introducing a KEMAR to

three diﬀerent positions along the x-axis 50, 75, and 100 cm to the left of

HATS (i.e. at 270

◦

). The results are presented in Figure 3.11 by the black,

blue, and red lines. The data clearly demonstrated that as an obstacle, the

second listener (KEMAR) does not inﬂuence the interaural time diﬀerence

when using Ambisonics, and HATS is at the center of the loudspeaker ring.

Chapter 3. Results 76

Figure 3.11: a) HATS alone at center. b) Light blue line: HATS alone at center.

Black line: HATS centered and KEMAR at 0.5 m to the right. Blue line: HATS

centered and KEMAR at 0.75 m to the right. purple line: HATS centered and

KEMAR at 1 m to the right.

3.3.2.2 Centered ILD

The eﬀects in higher frequencies due to a second listener require an analysis

of a diﬀerent parameter. Instead of studying the diﬀerence in the arrival time

of the sound between the ears, the representative metric is the level diﬀerence

between the ears.

There are eﬀects as absorption, reﬂection, and diﬀraction before the sound

pressure signal reaches the eardrums. The torso, shoulders, outer ear, and

pinna mechanically aﬀect an incoming sound wave. These eﬀects are angle

and frequency-dependent, as diﬀerent frequency waves have diﬀerent wave-

lengths [39, 40, 90].

The eﬀects on ILD caused by the virtualization process were calculated as the

diﬀerences between the reference ILDs measures with HATS alone and centered

and the ILDs measured with HATS and a second mannequin (KEMAR). As a

reference, Figure 3.12 presents the ILDs by each method from twelve diﬀerent

angles (30 degrees separation) around the listener.

Chapter 3. Results 77

Figure 3.12: Interaural Level Diﬀerences as a function of octave-band center fre-

quencies in twelve diﬀerent angles around the central point.

There are diﬀerences between ILDs calculated from measurement with both

techniques on the energy in the averaged octave bands. However, the ILDs

from VBAP present a signiﬁcant eﬀect based on incidence angle (more natu-

ral) than the Ambisonics [222]. Furthermore, the ILD peak for the Ambisonics

is observed around 2000 Hertz, which can be interpreted as the limit in fre-

quency reproduction of level diﬀerence between ears when decoding through

24 loudspeakers [299]. A more comprehensive comparison between techniques

with the HATS centered alone can be observed in the heatmap representation

from Figure 3.13 including all 72 angles (5 degrees separation) measured. The

homogeneity across angles from Ambisonics measurements indicates that its

ILD lacks precision as a binaural spatial cue. Localization accuracy in Am-

bisonics reproduction, especially to lateral angles, is highly dependent on its

order (acquisition and reproduction) [27].

Figure 3.14 shows the energy diﬀerence across the octave bands for eight dif-

ferent incidence angles on both techniques with and without the presence of

the second mannequin. On both techniques, the strongest inﬂuence happens

Chapter 3. Results 78

Figure 3.13: Interaural level diﬀerences averaged octave bands as a function of

azimuth angle for a HATS Br¨uel and Kjær TYPE 4128-C in the horizontal plane.

when the second mannequin is closest to the center. The second listener is to

the right angle in VBAP (270

◦

), while in the Ambisonics is positioned to the

left (90

◦

Figure 3.14: Interaural Level Diﬀerences (octave band) angles around the central

point considering diﬀerent displacement of the second listener.

The ILDs calculated from measurements with the second mannequin present

Chapter 3. Results 79

are not extensively diﬀerent compared to the reference ILD. The diﬀerence is

proposed to be observed as a distortion parameter. These diﬀerences were cal-

culated by subtracting the ILDs with the second mannequin from the speciﬁed

center alone reference ILD. Ideally all graphs should be black for a full match

(no diﬀerence between diﬀerent setups/positions), meaning no measured dis-

tortion.

Vector Based Amplitude Panning Figure 3.15 presents the diﬀerences

between ILDs calculated from HATS centered (HC) and the conﬁguration that

combines the HATS centered plus the KEMAR in one of the three positions

(e.g., HC K-50 is the deﬁned notation to HATS centered and KEMAR at 50

cm to the right). The sounds were auralized via VBAP for all 72 angles (5

◦

spacing). The angles that correspond to loudspeaker locations (15

◦

spacing)

is placed were reproduced directly by the physical loudspeaker at that angle.

Figure 3.15: VBAP discrepancies in ILD between HATS at the center and: (Top)

HATS at the center plus KEMAR at 50 cm to the right, (Middle) HATS at the center

plus KEMAR at 75 cm to the right, (Bottom) HATS at the center plus KEMAR at

100 cm to the right.

The diﬀerences in frequencies over 1 kHz are pronounced for angles to the right

side of centered HATS, 270-305

◦

azimuth. Smaller eﬀects can also be noted

Chapter 3. Results 80

on other angles that correspond to virtual sound sources (where there is no

loudspeaker, and the sound source is produced via the auralization technique).

These eﬀects are diminished as the second mannequin position increases away

from the centered receptor, indicating a smaller acoustic shadow.

Figure 3.16: VBAP Interaural Level Diﬀerences as function of azimuth angle

around the centered listener.

Figure 3.16 shows the ILD in six octave bands from impulse responses recorded

with ﬁles auralized using VBAP. The HATS centered (HC) position refers to

HATS alone and it is compared to the conﬁgurations adding the second listener

(KEMAR) in three diﬀerent positions 50, 75 and 100 cm displaced from the

center (K+50, K+75, and K+100, respectively).

The mismatch is pronounced when KEMAR is closer (blue line), especially

in the angles blocked by KEMAR. As the second listener blocks the sound

wave, an acoustic shadow is created, which reduces the sound energy to the

ear facing the sound source, decreasing the level diﬀerence between ears. There

is also a reduction in ILD for angles from 35 to 50. That can be related to the

opposite eﬀect where the mannequin reﬂects part of the sound, increasing the

level to HATS-centered counter ear. The ﬁnding supports interpreting that a

Chapter 3. Results 81

substantial eﬀect occurs on ILDs to the KEMAR’s closer position.

Ambisonics Figure 3.17 presents the calculated diﬀerences between ILDs

from the Ambisonics auralization with the same conﬁgurations (i.e., HC vs.

HC K-50, HC vs. HC K-75, and HC vs. HC K-100).For convenience, the second

mannequin was positioned to the left of the center (90

◦

). The switch from right

to left does not aﬀect the comparison as both HATS and the Eriksholm test

room are symmetric. Figure 3.18 shows the ILD in octave bands, to highlight

the stronger eﬀect being at the 8 kHz.

Figure 3.17: Ambisonics discrepancies in ILD between HATS at the center and:

(Top) HATS at the center plus KEMAR at 50 cm to the left, (Middle) HATS at

the center plus KEMAR at 75 cm to the left, (Bottom) HATS at the center plus

KEMAR at 100 cm to the left.

The results demonstrate that including a second listener has a negligible ef-

fect on Ambisonics ﬁrst-order ILDs. However, on the reference measurement

(HATS alone), the ILDs did not adequately reproduce this spatial cue through-

out the angles around the listener, given the observable minor ILD diﬀerences

across angles, especially over 2 kHz.

Chapter 3. Results 82

Figure 3.18: Ambisonics Interaural Level Diﬀerences as function of azimuth angle

around the centered listener.

3.3.3 Oﬀ-centered position

Being able to have the participant away from the center of the loudspeak-

ers ring can be valuable for testing simultaneous participants or a particular

physical apparatus’ inﬂuence (e.g. Listening eﬀort evaluated under presence

of another individual [230]). Auditory research that aims to test the inﬂuence

of a particular noise, SNR, or the direction of the noise on the interaction in

participants’ conversation can beneﬁt from a setup that would make it possible

to virtualize a sound scene and present it without spatial distortions. Mea-

surements aiming to study the inﬂuence of oﬀ-center HATS displacement were

performed in nine diﬀerent conﬁgurations: with HATS and KEMAR indepen-

dently displaced 25, 50, and 75 cm from the center, resulting in separations of

50, 75, 100, 125, and 150 centimeters (See Figure 3.6b).

The listening position is critical to the auralization process techniques pre-

sented in this work as they are derived and programmed to render the sound

in the center of a loudspeaker array. Adding computer power to real-time pro-

Chapter 3. Results 83

cessing could handle participant movements; although that can be considered,

it was not in this part of the experiment scope. Such processing focuses on

dynamics (head motion). The focus here is the eﬀects of sub-optimal positions

and the inﬂuence of a second listener as an obstacle to the sound ﬁeld.

3.3.3.1 Oﬀ-center ITD

The eﬀects of oﬀ-center positioning on sound’s arrival time can aﬀect the sub-

jective perception of the sound incidence direction.

Vector Based Amplitude Panning Observing the ITD results shown in

Figures 3.19, 3.20, and 3.21, almost no inﬂuence of the second mannequin

(KEMAR) can be noted even with the HATS oﬀ-center. The ITD at oﬀ-center

positions deviates from the ITD from HATS centered at the same proportion

regardless the second listener (KEMAR) position.

Nonetheless, Figure 3.22 shows that a pronounced eﬀect appears by shifting out

the HATS oﬀ center. When exceeding 25 cm, the spikes represent a diﬃculty

of the vector-based amplitude panning process to generate the virtual sound

sources. This behavior is expected as the VBAP mathematical formulation is

derived by a unitary vector pointing to the center.

In ﬁgures 3.20 and 3.21 it is possible to observe more considerable distortions

(sharp peaks crossing the reference line in addition to being oﬀset from the

reference line) in the ITD for the virtual sound sources. Such distortions

increase as HATS is moved away from the central position. Sound sources

reproduced using VBAP in this loudspeaker ring to these receptor positions

would not be correctly interpreted in terms of direction by the listener.

The ITDs diﬀerence is greater when the sound sources are at angles close to the

Chapter 3. Results 84

Figure 3.19: ITD as a function of source angle Light blue line: HATS alone at

the center. Black line: HATS at -25, KEMAR at +25. Blue line: HATS at -25,

KEMAR at +50. Red line: HATS at -50, KEMAR at +75.

Figure 3.20: ITD as a function of source angle Light blue Line: HATS alone at

the center. Black Line: HATS at -50, KEMAR at +25. Blue line: HATS at -50,

KEMAR at +50. Red line: HATS at -50, KEMAR at +75.

front or rear (0

◦

and 180

◦

) directions. This eﬀect is related to HATS physical

displacement. The ITD results at lateral angles are representing a larger lobe

to the HATS right ear (270

◦

), and a sharpened lobe at the HATS left ear (90

◦

)

Chapter 3. Results 85

Figure 3.21: ITD as a function of source angle. Light blue line: HATS alone,

centered. Black line: HATS at -75, KEMAR at +25. Blue line: HATS at -75,

KEMAR at +50. Red line: HATS at -50, KEMAR at +75.

Figure 3.22: ITD as a function of source angle. Light blue line: HATS alone,

centered. Black line: HATS at -25, KEMAR at +25. Blue line: HATS at -50,

KEMAR at +50. Red line: HATS at -75, KEMAR at +75.

shows the oﬀ-center displacement. This eﬀect occurs because HATS is not

at the center of the ring (See Figure 3.23b), and the angles and separations

between the loudspeakers are modiﬁed. The eﬀect is even more apparent when

Chapter 3. Results 86

looking only at the real sound source (angles correspondent to loudspeaker

locations) ITDs, without the distortions created by VBAP auralization, (See

Figure 3.23a).

(a) (b)

Figure 3.23: a) ITD for real sound sources. Light blue line: HATS alone, centered.

Black line: HATS at -25, KEMAR at +25. Blue line: HATS at -50, KEMAR at

+50. Red line: HATS at -75, KEMAR at +75. b) HATS oﬀ-center position -75 cm

scheme facing the third loudspeaker.

Ambisonics The VBAP method constructs the auditory spatial cues through

one to three loudspeakers in this setup, usually in the same quadrant. Am-

bisonics, in contrast, uses all the available loudspeakers in the rendering pro-

cess. Hence, the sound localization is beneﬁted on VBAP auralization com-

pared to Ambisonics due to the nature of the methods [104, 105, 175, 180, 221].

Furthermore, the ITDs results observed from the ﬁrst-order Ambisonics reﬂect

the method’s limitation on sweet spot size.

Figure 3.24 shows the calculated ITD in three diﬀerent conﬁgurations H+25

K-25, H+50 K-50, H+75 K-75, and the center conﬁguration for comparison.

To improve readability, the ITD results for the remaining spatial conﬁguration

(which were similar across conditions) can be found in Appendix A.

The expected size of a listening area is 20 cm when combining 24 speakers

to reproduce Ambisonics in a 2D horizontal matrix [299]. The displacement

Chapter 3. Results 87

Figure 3.24: ITD as a function of source angle in Ambisonics setup. Light blue

line: HATS alone, centered. Black line: HATS at -25, KEMAR at +25. Blue line:

HATS at -50, KEMAR at +50. Red line: HATS at -75, KEMAR at +75.

of 25 cm and greater puts the receptor outside the sweet spot. Therefore,

it is possible to observe in Figure 3.24 that Ambisonics does not virtualize

this acoustic track correctly outside the center position, as the values remain

mostly constant for the side being played.

3.3.3.2 Oﬀ-center ILD

ILDs can be highly sensitive to the listener’s position on a virtualized sound

ﬁeld, given the considered smaller wavelengths. The composition of a virtu-

alized sound wave is be performed by simultaneously combining sounds from

several sound sources, which requires a highly precise combination.

This section investigates the ILD changes due to having the listener away from

the optimal position while having another listener present. ILD’s inﬂuence

when the HATS and a second participant are away from the center.

A comparison of the ILD results across the positions is shown in Figure 3.25;

it presents for both techniques the calculated ILDs over frequency on eight

Chapter 3. Results 88

incidence directions spaced over 45 degrees in the azimuth on three diﬀerent

positions plus the centered position as a reference. The pattern deviation as

the receptor is moved from the center is not the same across the techniques.

As expected, the physical construction of the summed sound wave from Am-

bisonics that relies on all loudspeaker has a higher impact on ILDs than the

VBAP which only combines few sound sources from the same quadrant.

Figure 3.25: ILD as a function of frequency at diﬀerent angles (line color) for

VBAP (top row) and Ambisonics (bottom row) for symmetrical displacement in oﬀ-

center setups.

On ﬁles auralized through VBAP, the discrepancies between ILD measured

having HATS in the center (optimal position) and the other positions can be

interpreted as acoustic artifacts capable of conveying the wrong localization

of the sound source. Although the second listener did not have a primary

inﬂuence, the observed displacement from the center aﬀects the ILD pattern,

especially the higher frequencies. For Ambisonics, the listener position is crit-

ical. The ILD diﬀerences from center to oﬀ-center positions create artifacts

that compromise ILD used as a cue to the sound localization on all tested

positions.

Chapter 3. Results 89

Vector Based Amplitude Panning The top row of Figure 3.25 shows the

ILD screening in some of the incidence angles. Comprehensive visualization

of ILDs across angles is presented in Figure 3.26 for the reference-centered

(top) and oﬀ-centered positions. There is an eﬀect on ILDs when moving

the receptor from the center position and adding a second listener inside the

loudspeaker’s ring. Although noticeable, the eﬀect still preserves the pattern,

allowing the diﬀerence to be interpreted as artifacts. The vertical zeroes ILDs

indicated the frontal and rear angles (0

◦

and 180

◦

) where the sound should

arrive at the ears with the same level. These vertical black lines are shifted

as the listener is displaced from the center. At 75 cm displacement the lowest

value vertical line on Figure 3.26 appears is 35

◦

(frontal) and 145

◦

(rear)

Figure 3.26: VBAP setups: ILD on centered position (top); ILD on oﬀ-center

setups: HATS at 25 cm to the left with KEMAR at 25 cm to the right (middle top);

HATS at 50 cm to the left with KEMAR at 50 cm to the right (middle bottom);

HATS at 75 cm to the left with KEMAR 75 cm to the right (bottom).

The diﬀerence between ILD with the HATS in the reference position (alone

and in the center) and the conﬁgurations with HATS outside the center simul-

taneously with KEMAR are shown in Figures 3.27, 3.28 and 3.29.

The acoustic ﬁeld behavior outside the center of the ring at frequencies above

Chapter 3. Results 90

Figure 3.27: VBAP diﬀerences in the ILD between centered Alone and oﬀ-center

with KEMAR setups: HATS at 25 cm to the left with: KEMAR at 25 cm to the

right (top); KEMAR at 50 cm to the right (middle); KEMAR 75 cm to the right

(bottom).

Figure 3.28: VBAP diﬀerences in the ILD between centered setup and 25 cm oﬀ-

center VBAP setups: HATS at 50 cm to the left with: KEMAR at 25 cm to the

right (top); KEMAR at 50 cm to the right (middle); KEMAR 75 cm to the right

(bottom).

1 kHz presents signiﬁcant ILD diﬀerences for the measured conﬁgurations,

especially on angles that are virtual sound sources. The ILD diﬀerence reaches

up to 15 dB.

As in the ITD, the ILD data from HATS in the oﬀ-center position shows the

acoustic shadowing eﬀect caused by KEMAR. It is possible to note that as

Chapter 3. Discussion 91

Figure 3.29: VBAP diﬀerences in the ILD between centered setup and oﬀ-center

setups: HATS at 75 cm to the left with: KEMAR at 25 cm to the right (top);

KEMAR at 50 cm to the right (middle); KEMAR 75 cm to the right (bottom).

close as KEMAR is positioned to HATS, greater discrepancies in ILD around

positions near 270 degrees occur. This eﬀect is due to the diﬀraction and

absorption of the sound on the second listener (KEMAR), and happens for

both real (loudspeaker) and virtual sound sources locations.

Ambisonics Ambisonics presents a more considerable limitation regarding

movement outside the center of the ring due to its nature. The sound compo-

sition requires a combination of amplitude and phase from all available loud-

speakers being the correct representation achieved only for an area at the center

and without obstructions. The ILD in octave bands is shown in Figure 3.30.

The low amplitude and homogeneity across frequencies demonstrate that Am-

bisonics is limited to render the binaural cue proposed, not appropriately de-

livering the level diﬀerences outside the center. The ILD diﬀerences from the

oﬀ-center positions to the HATS centered are presented in Appendix B.

Chapter 3. Discussion 92

Figure 3.30: Ambisonics setups: ILD on centered position (top); ILD on oﬀ-center

setups: HATS at 25 cm to the left with KEMAR at 25 cm to the right (middle top);

HATS at 50 cm to the left with KEMAR at 50 cm to the right (middle bottom);

HATS at 75 cm to the left with KEMAR 75 cm to the right (bottom).

3.4 Discussion

Once the listener is centered in the loudspeaker array, the second listener did

not aﬀect the auralization other than the angles physically shadowed by the

second listener. Thus, a second listener does not deteriorate the spatial cues

on both auralization techniques analyzed in this work.

For VBAP, the discrepancies in ITD only occur as the second listener was

positioned 50 cm away, the closest measured position in this experiment. Also,

diﬀerences in ILD for VBAP are more notable at the second listener’s closest

position. Concurrently, Ambisonics has not presented an apparent diﬀerence

in ITD to a centered listener by placing a second listener inside the ring.

The diﬀerence in Ambisonics ILDs from the centered reference indicates an

acoustic shadow (this time at the left angle of 90 degrees) and an additional

slight diﬀerence across other angles.

Chapter 3. Discussion 93

There is an apparent eﬀect on ITD as the listener is moved out of the center.

For VBAP, the peak of magnitude remains practically the same, approximately

650 microseconds, as the ITD 0 value (sound reaching simultaneously in both

ears) is shifted. At 75 cm oﬀ-center to its left side, the diﬀerence in arrival time

corresponds to a shift of 30 degrees approximately. That is in line with the

setup, as the mannequin was placed in front of another loudspeaker. However,

the Ambisonics ITDs demonstrate that the composition of magnitude and

phase is not completed in oﬀ-centered positions. The Ambisonics weights are

calculated to the sound waves from the loudspeakers to interact in the center

position and then form a sound ﬁeld representing a sound wave from a deﬁned

incidence angle. Moving the primary listener to the right makes the interaction

between the loudspeakers inaccurate. In this case, the time diﬀerence turns

wrong due to the Ambisonics truncation order to be low increasing the aliasing

eﬀect as it can be similarly observed in third and ﬁfth order in Laurent et

al., [275].

The sound from the right mainly reaches the right ear and travels to the left

ear before the sound from the left side can travel the extra distance. That is

an expected eﬀect since even the minimum displacement (25 cm) is larger than

the expected reproducible area (around 20 cm) for this setup. There was no

diﬀerence observed as the second listener (KEMAR) positions were changed

(25, 50, and 75 cm to the right of the center) in all VBAP measurements

with HATS positioned to the left of the center. Considering the ITD just

noticeable diﬀerence (JND) in an anechoic condition is of the order of the 10

to 20 microseconds [38, 140, 241], the ITD results when the oﬀ-center HATS

position was 25 cm to the left were a good approximation of the reference-

centered measurement.

That means that a listener would not be able to discern the diﬀerence con-

cerning the direction of incidence if placed in these positions relying only on

Chapter 3. Discussion 94

the ITD cue. It is also worth considering that the JND to reverberant condi-

tions is even higher [140] and the artifact can be masked by reverberation [97]

which would beneﬁt the auralization process. The HATS measurements posi-

tioned on 50- and 75-cm presents peaks and crossover values across the line

that corresponds to centered ITD, which indicate distortion problems at low

frequencies regarding the spatial cue. A similar analysis of the KEMAR im-

pact on ITDs from Ambisonics virtualization can not be achieved since the

ITD is not accurately rendered outside of the sweet spot.

Each result of the interaural level diﬀerence position combination (HATS and

KEMAR) was subtracted from the HATS results alone to perform the ILD

analysis oﬀ-center. In the VBAP method, a shadow eﬀect generated by a

second listener is present as expected, mainly when the ﬁrst listener is 25 or 50

cm left of center. However, the diﬀerences in high frequencies are essentially

on virtual sources, which indicates the diﬃculty of creating the virtual sound

source impression outside the center position, independently of the second

listener presence [2].

Oﬀ-center positions did not allow the accurate synthesis of the ILDs from the

loudspeakers using Ambisonics. The method did not reproduce time or level

diﬀerences accurately in these conditions, which could lead to not achieving the

correct spatial impression. That is in line with literature, although generally

investigating higher Ambisonics orders, the complexity of accurately render

high frequency cues is present [279, 290], and also the oﬀ-center increase on

accuracy by increasing the Ambisonics order with a proper number of loud-

speakers [275].

It should be noted that the current study did not measure changes in ITD and

ILD for oﬀ-center listener positions without the presence of a second listener.

Based on the eﬀects of having the ﬁrst listener oﬀ-center with a second listener

Chapter 3. Concluding Remarks 95

present, coupled with the smaller changes with a second listener when the ﬁrst

listener is centered, it can be deduced from the current results that the oﬀ-

center position has a degradation eﬀect on the ITD and ILD. Considering that

many simulations are limited by a “sweet spot” for the listener(s), the oﬀ-

center position, as opposed to the presence of a second listener, is probably

the greatest liability for multi-listener methods in hearing research.

3.5 Concluding Remarks

The more demanding the test requirement in terms of localization of the sound

source (out from left, right, front and back), the more the researcher should

drive towards VBAP. In case of ﬁxed positions and requirement of more sense

of immersion, Ambisonics should be able to build more convincing sound sce-

narios.

The techniques do not aﬀect the ILD and ITD acoustic cues in the central

position for one test participant. The addition of a second listener within the

ring also does not signiﬁcantly aﬀect these parameters at the three distances

tested, except for the angles usually hidden by the shadow second listener.

Thus, it is suitable to move towards subjective tests with a center participant

and an actor on the side. Although the second listener has not deteriorated

the techniques, they present diﬀerent performances in terms of spatial repre-

sentation and notably present a diﬀerent sense of immersion. Thus, the test’s

purpose to be designed must be taken into account when deﬁning the aural-

ization method.

There is a clear degradation when two test subjects are simultaneously present

both in oﬀ-center positions, regardless of the distance of a second listener. The

VBAP measurements showed an increase in diﬀerences for ITD increasing the

Chapter 3. Concluding Remarks 96

distance from the center and signiﬁcant diﬀerences in ILD. These diﬀerences

indicate the creation of acoustic artifacts, possibly generated by the method’s

diﬃculty in correctly virtualizing high frequencies outside the sweet spot. For

the ITD parameter, the displaced position of 25 cm of the center has little

diﬀerence or evidence of artifacts generated by virtualization errors. At the

same time, the other distances present signiﬁcant diﬀerences and artifacts. The

binaural cues analysis suggests that VBAP is less sensitive to the participant

positions than the Ambisonics setup.

However, it is relevant to note that although the diﬀerences in the binaural

cues denote diﬀerences in audio spatialization, reﬂecting on the perceived angle

of incidence of the sound, both techniques can be calibrated to reproduce the

stimuli at a desired level of sound pressure. That means that an auralized

sound can be reproduced with the correct sound pressure level although its

direction may not be correctly interpreted by the listener as their binaural

cues are not being delivered appropriately.

Chapter 4

Subjective Eﬀort within

Virtualized Sound Scenarios

This experiment was a collaborative study (EcoEG [3]) with fellow HEAR-

ECO PhD student Tirdad Seiﬁ-Ala, also from the University of Nottingham,

that combined the virtualization of sound sources and electroencephalography

(EEG) to assess listening eﬀort in ecologically valid conditions. Both students

contributed equally to the study design, preparation, data collection and in-

terpretation. TSA additionally performed the data analysis; SA additionally

performed the room simulations, stimuli preparation, software interface and

sound calibration.

As deﬁnitions can vary, this chapter uses the following terms:

• Simulation: Numerical acoustic simulation of spatial behavior of a sound

in a deﬁned space.

• Auralization: Creation of a ﬁle that can be converted to a perceivable

sound and contains spatial information.

Chapter 4. Introduction 98

• Sound Virtualization: Reproduction of an auralized sound ﬁle through

loudspeakers or headphones.

4.1 Introduction

The interest from researchers and clinicians in the listening eﬀort measures has

grown recently [83, 135, 210], the importance of studying listening eﬀort in an

ecologically valid sound environment follows the same trend [134].

The previous chapter discussed the feasibility and constraints of the virtual-

ized sound ﬁeld through binaural cues and foreseeable eﬀects on spatial im-

pression and localization. This chapter investigates whether reverberation and

the signal-to-noise ratio (SNR) are modeled in behavioral data, being a proxy

of subjective listening eﬀort in a virtualized sound environment.

The reverberation is the accumulation of energy reﬂections (sound) in an en-

closed space that creates diﬀusion in its sound ﬁeld [256]. Reverberation Time,

in turn, is an objective parameter that represents the amount of time required

to dissipate the energy of a sound source by one-millionth of its value (60 dB)

after the sound source has ceased [254]. This parameter was reviewed in Sec-

tion 2.3.3.2. The remaining sound energy can blur the auditory cues, rapid

transitions between phonemes, and decrease the low-frequency modulation of

a signal; it may compromise speech intelligibility [39, 112].

Since reverberation is a complex phenomenon, depending on space and fre-

quency [111, 185], a wide range of physical-acoustical factors may limit some

comparisons. For example, the reproduction method, the masker type, the

position and number of sources, the SNR, the sound pressure level of the pre-

sentation, the reverberation time interval studied, and the simulated position

being in a free or in a diﬀuse sound ﬁeld. Like the methodologies, the ﬁndings

Chapter 4. Introduction 99

in terms of reverberation inﬂuence on listening eﬀort across experiments can

also vary.

Previous studies investigated the eﬀect of reverberation on speech intelligi-

bility and listening eﬀort. Variations across reverberation time, level, and

population groups were observed. For example, a correlation between age and

reverberation was traced in work by Neuman et al. [203]. This study found that

reverberation negatively impacts the necessary SNR to reach 50% of speech

recognition. This impact varies across ages, with the eﬀect decreasing as age

increases. The sensitivity of subjective measures and electrodermal activities

were evaluated by Holube et al. [121]. The eﬀect of reverberation was found

statistically signiﬁcant to subjective measures but not to the electrodermal

activity. A study from Picou et al. [225] presents a response time in a dual-

task paradigm as a behavioral measure of listening eﬀort. In their study, there

was no signiﬁcant eﬀect in response time neither in the same SNR conditions

nor comparing the response time of equal performance scores. The impact on

listening eﬀort was studied by Kwak et al. [149] through subjective ratings

resulting in a signiﬁcant eﬀect of reverberation on ratings of listening eﬀort

and the sentence recognition performance. In Nicola and Chiara’s study [204],

the negative inﬂuence of reverberation on response time was considered in-

dicative of an increase in listening eﬀort. The study assessed the inﬂuence of

reverberation and noise ﬂuctuation on response time. The diﬀerent method-

ologies applied in the studies and their groups of participants must be carefully

analyzed, as they can explain the diﬀerent results.

Ambisonics arrangements (Mixed Order Ambisonics (MOA) [78, 177] and

HOA) are already used in audiological studies [7, 77, 173, 303]. This study

proposed a low-order (ﬁrst-order) Ambisonics implementation. The low-order

technique is more sensitive to the listener position [64, 65], which was also veri-

ﬁed in this study. That can be seen as a counter-intuitive and non-conventional

Chapter 4. Methods 100

choice, although it was meant to assess low-order Ambisonics’ feasibility in

audiological studies and its constraints. This decision was a step towards con-

ﬁrming the feasibility of a listener in a centralized position found in Chapter 3,

observing its constrains, and further developing an auralization method with

lower hardware requirements in Chapter 5.

Hypothesis The main research question is how the auralized acoustic sce-

nario, speciﬁcally the room and the SNR, increases auditory eﬀort when vir-

tualized. The hypothesis for the experiment is that a longer RT provided

through sound virtualization and a lower SNR both lead to a more signiﬁcant

listening eﬀort. Reverberation time can inﬂuence normal hearing and hearing-

impaired people in diﬀerent ways. For example, on average, hearing-impaired

listeners experience more signiﬁcant diﬃculties with understanding speech in

a reverberant condition than normal hearing listeners, so they can suﬀer more

from the strain of listening. As reverberation’s eﬀects on hearing-impaired

listeners vary (see Chapter 2), this study employed only normal-hearing par-

ticipants to investigate the eﬀects of audio degradation. To subjectively assess

changes in hearing eﬀort, a questionnaire was provided to participants, asking

how much eﬀort they found for each condition (described in Section 4.2). This

investigation is the ﬁrst step to understanding the feasibility of including the

simpliﬁed virtualization of sound sources in the expanding ﬁeld of listening

eﬀort research.

4.2 Methods

This experiment was designed to gather data for two parallel analyses: the ﬁrst

was to evaluate diﬀerences in behavioral performance (speech recognition) and

subjective impressions of listening eﬀort driven by diﬀerent scenarios, manip-

Chapter 4. Methods 101

ulating the room type and the signal-to-noise ratio (SNR). The second study

compared physiological responses of the brain as measures of listening eﬀort

to the same behavioral performance. This chapter focuses on the experiment’s

ﬁrst study (behavioral data vs. subjective impressions). Three rooms were

chosen for this study: a classroom, a restaurant dining area, and an anechoic

room.

For this experiment, a setup was developed to investigate the inﬂuence of lis-

tening eﬀort caused in nine diﬀerent situations: three room simulations charac-

terized by their reverberation time and three SNRs. The setup was composed

of four recorded talkers acting as maskers and one talker acting as the target.

The talkers’ positions were all spatially separated.

The test paradigm involved the auditory presentation of Danish hearing in

noise test (HINT) sentences [205] on top of four speech maskers and recalling

the words they could keep in memory after 2 seconds. The sound sources are

spatially distributed and the participant is informed that the target speech

is always from the front. The participants responses were word scored (i.e.,

word-based speech intelligibility) by Danish-speaking clinicians.

The method in this study follows a similar setup with a four-talker babble setup

as in [209, 302], which investigated SNR and masker types using pupilometry

as a proxy for listening eﬀort. Also, a study from Wendt et al., [301] inves-

tigated the impact of noise and noise reduction through an equivalent setup.

This method’s innovation relies on using ﬁrst-order Ambisonics to generate the

reverberation based on Odeon simulated rooms.

Chapter 4. Methods 102

4.2.1 Participants

For the data collection, 18 normal-hearing native Danish-speaking adults (eight

females) with an average age of 36.9 ± 11.2 years ﬁrst gave written consent form

and initially participated in the test. One participant was placed outside the

sound ﬁeld sweet spot, so his data were discarded, and the data for the other

17 participants were used for further analysis. Ethical approval for the study

was obtained from the Research Ethics Committees of the Capital Region

of Denmark. For each participant, the pure-tone average of air conduction

thresholds at 0.5, 1, 2 and 4 kHz pure tone audiometry (PTA4) were tested

and conﬁrmed below 25 dB HL.

4.2.2 Stimuli

The target stimulus consisted of simple Danish sentences spoken by a male

speaker. The sentences were from the HINT in Danish [205] and were 1.3-1.8

s in duration. The masking signal consisted of four diﬀerent speakers, two

female and two male, reading a Danish-language newspaper [302]. The total

duration of each of the masker recordings was approximately 90 seconds. The

maskers’ onset was 3 s before and oﬀset was 2 s after the target, resulting in

a masker duration of 6.3-6.8 s. In each trial, the time segment used of each

masker was randomized. In addition, the spatial position for each masker was

also randomized in each trial, but always interspersing male and female talkers.

The overall maskers’ equivalent continuous sound level L

was set at 70 dB

(64 dB each masker), and the target L

were set at 62 dB, 67 dB and 72 dB

to generate three diﬀerent SNR conditions: -8, -3 and +2 dB. In this study,

SNR was deﬁned as the equivalent continuous sound level of the target signal

compared to the competing masking L

. The chosen reverberation conditions

aimed to represent common everyday situations. The RT of the anechoic and

Chapter 4. Methods 103

reverberant conditions studied were deﬁned as the overall reverberation time

obtained through the output of the simulation software (ODEON Software

v.12). The chosen reverberation time values aim to represent common everyday

situations. The absorption coeﬃcients and relative area utilized to obtain the

mentioned conditions are presented on Appendix E.

Five source positions (one target and four maskers) were created around a

receptor in each simulated room. All positions were 1.35 m from the center

of each room where the receptor is located. The approach of creating two

diﬀerent rooms instead of changing the parameters of a single room was chosen

to achieve a more natural sound ﬁeld. That way the absorption coeﬃcient

applied to the room’s materials was kept close to real.

The virtualization of the proposed acoustic scenarios follows the path indicated

in Figure 4.1. An acoustic simulation is performed to create the appropriate

characteristics of the sound according to the room. The software calculates

the amplitude and the incidence directions of sound and its reﬂections arriving

from speciﬁc sources to a receptor position inside the room. For each combina-

tion of source-receptor, the software generates a room impulse response that is

encoded in Ambisonics ﬁrst-order in AmbiX [198] format (which is a channel

order speciﬁcation for Ambisonics auralization ﬁrst 4 channels are WYZX com-

pared to WXYZ in the FuMa speciﬁcation). The generated ﬁle was convolved

with anechoic audio and decoded to the speciﬁc array of 24 loudspeakers.

Chapter 4. Methods 104

Figure 4.1: Auralization procedure implemented to create mixed audible HINT

sentences with 4 spatially separated talkers at the sides and back (maskers) and one

target in front.

4.2.3 Apparatus

The experiment was set up in an anechoic room (IAC Acoustics) with 4.3 m

× 3.4 m × 2.7 m (inner dimensions). The experimental setup consisted of

a circular array of 24 loudspeakers positioned on 15

◦

interval on the azimuth

and 1.35 meters distance from the center. The target sound was reproduced

at 0

◦

(participant’s front), the maskers were auralized at ± 90

◦

and ± 150

◦

(Figure 4.2). The position of the participant during all the test was monitored

through a laser line and a camera ensuring they remained in the sweet spot.

Stimuli were routed through a sound card (MOTU PCIe-424) with Firewire

440 connection to the MOTU Audio 24 I/O interface) and were played via 16

loudspeakers Genelec 8030A and 8 loudspeakers Genelec 8030C (Genelec Oy,

Iisalmi, Finland) aligned in frequency and level. The Biosemi EEG device was

used to collect the physiological data, which helped to restrain participants’

movement; the EEG data were not analyzed in this study.

Chapter 4. Methods 105

Figure 4.2: Spatial setup of the experiment: Test subjects attended to target (in

blue) stimuli from a 0

◦

angle in front.The masking talkers (in red) are presented at

lateral ±90 and rear ±150 positions.

All enclosed spaces have a certain degree of reverberation due to acoustically

reﬂective surfaces and background noise due to equipment, including controlled

audiological environments. The levels of reverberation and background noise

meet the criteria from Recommendation ITU-R BS.1116-3 [126] and are re-

spectively shown in Figures 4.4 and 4.3.

Figure 4.3: Reverberation Time inside anechoic room at Eriksholm Research Cen-

tre with setup in place.

Chapter 4. Methods 106

Figure 4.4: Eriksholm Anechoic Room: Background noise A-weighted. Loudspeak-

ers and lights on, motorized chair oﬀ.

The parameters were measured with the setup (loudspeakers, motorized chair

and BioSemi eeg equipment) inside the room and positioned as in the experi-

ment. Figure 4.5 shows the setup placed inside the anechoic room.

Figure 4.5: Setup inside anechoic room (Motorized chair, adjustable neck support

and EEG equipment).

Chapter 4. Methods 107

4.2.4 Auralization

Acoustic Scene Generation and Room Acoustic Simulation

To simulate the acoustics characteristics of the chosen scenarios, geometric

models were created in the room acoustics software ODEON. Next, the Am-

bisonics Room Impulse Responses were simulated using ODEON software,

version 12 [59]. The absorption coeﬃcients of the room surfaces are listed in

Annex E. All sentences were auralized in Ambisonics [15], truncated by 1st

order and encoded to 24 channels. The analysis utilized the Institute of Tech-

nical Acoustics (ITA)Toolbox [29, 67]. Rooms were chosen as representative

of realistic and not extreme acoustic conditions. The spaces simulated were a

classroom (9.46 m × 6.69 m × 3.00 m) with an overall RT of 0.5 seconds, and

a restaurant’s dining area (12.19 m × 7.71 m × 2.80 m) with an overall RT

of 1.1 seconds. The distance between source and receptor was kept the same,

1.35 m, across rooms. Target and masker positions were simulated by selecting

the appropriate simulated RIR to convolve i.e., the simulated source-receptor

RIR that corresponds to the desired reproduction angle.

Ambisonics Sweet Spot In this study, two diﬀerent metrics were used

to compare the oﬀ-center performance of virtual sources auralized with ﬁrst-

order Ambisonics: the RT and the Sound Pressure Level (SPL). That is, the

presented virtualized soundﬁeld was delivering the correct amount of reverber-

ation and also the correct sound pressure level of each source resulting in the

appropriate signal-to-noise ratio when was not perfectly centered. To estimate

each position’s metrics, a logarithimc sweep signal (50-20000 Hz, 2.73 s (FFT

Degree 18, Sample Frequency 96 kHz)) was generated and convolved with the

Ambisonics ﬁrst-order RIR calculated by ray-tracing in ODEON for each mod-

eled room. The simulated rooms presented an overall theoretical reverberation

Chapter 4. Methods 108

time of 0, 0.5, and 1.1 s. These auralized ﬁles were encoded to 24 channels

distributed in the horizontal axis. In the following, the ﬁles were played in-

side the anechoic room and simultaneously recorded. From the division in the

frequency domain of the recorded signal and the zero-padded initial signal (de-

convolution), the calculated impulse response (or binaural RIR (BRIR) when

recorded with HATS) represents the virtualized system, including the physical

eﬀects of the array and all calibration.

Reverberation Time The RT was calculated with ITA-Toolbox from ini-

tial 20-dB decrease from peak level (T

) in the virtualized IRs. Figure 4.6

shows the overall RT results at the center position and by moving the receptor

(manikin) towards the front.

Figure 4.6: Overall reverberation time (RT) as a function of receptor (head) posi-

tion in the mid-saggital plane re center (0 cm)

The results showed slightly greater RTs (0.58 and 1.16 s) than what was sim-

ulated in the ODEON software (0.5 and 1.1 s). However, this was expected

since there is equipment inside the anechoic room (e.g., a large chair and loud-

speakers) that can be considered reﬂective surfaces that were not present in

Chapter 4. Methods 109

the simulation. The results showed that there is no major eﬀect on the energy

decay for small head movements.

Sound Pressure level The sound pressure level was determined by con-

volving the target and masker sounds with the impulse responses collected

across twelve positions with horizontal displacements of 2.5, 5 and 10 cm and

forward (mid-saggital) displacements of 2.5 and 5 cm. The results are shown

in Figure 4.7. Four speech talkers are individually convoluted. The equivalent

sound pressure level is determined using the calibration factor. The measure

is the average of 20 diﬀerent sentences.

Figure 4.7: Sound pressure level virtualized through Ambisonics at diﬀerent listener

positions.

The changes in SPL as a function of oﬀ-centre position do not follow a consis-

tent pattern. The SPL changes were, however, mostly similar across the three

simulated rooms, with the exception of three positions where the restaurant

(1.1 s RT) was 1-1.5 dB diﬀerent (x = 2.5, y = 0; x = 0, y = 2.5; x = 10;

y = 2.5). The center position is the optimal position for sound pressure level

accuracy. To help get reliable, appropriate data from the experiment, a neck

rest as well as a video feed and laser line were added to the setup after the

Chapter 4. Methods 110

ﬁrst pilot test. The participants were asked to be in contact with the neck rest

all the time. The clinician was able to see the laser line at the patient’s head

throughout the test. They could ask the participant to quickly correct posture

at the start of each block or at any point of the session after the participant

needed a break. Figure 4.8 shows a participant positioned with all sensors con-

nected. Another important ﬁnd was, after adjusting the participant position,

the motorized chair should be unplugged, otherwise the EEG data would be

compromised.

Figure 4.8: Participant positioned to the test.

4.2.5 Procedure

There were 9 diﬀerent conditions based on SNR (+2 dB, -3 dB, -8 dB) and

reverberation time (0 s, 0.5 s, 1.1 s) of the sound. Each condition was presented

in separate blocks, and each block consists of 20 sentences, so in total there

were 9 blocks and 180 sentences presented to the participants in the main

test. In addition to that, each participant went through a training round

in the beginning, consisting of 20 sentences with diﬀerent conditions. The

procedure for each trial is illustrated in Figure 4.9. Each trial started with 2 s

Chapter 4. Methods 111

of silence (preparation), then 3 s of background noise which served primarily

as a baseline period for the separate EEG analysis. Then a HINT sentence

was played as the background noise continued for 1.5 s on average. After

the target sentence ﬁnished, the background noise continued for another 2

seconds during which participants needed to maintain the words they just

listened to (maintenance), also serving primarily for the companion analysis

of EEG responses re baseline. When the background noise was stopped, the

participants were instructed to repeat all the words within the sentence (recall).

The listening eﬀort reﬂected in alpha power changes in the maintenance phase

have been investigated by [208, 310, 311, 313]

Figure 4.9: Trial design. For each trial, 20 in each block, there was 2 s of silence,

then 3 s of masker (4 spatially separated talkers), then a Danish HINT sentence as

target stimuli in the presence of continuing masker, then 2 additional s of masker,

followed by silence when the participant repeated as many target words as they could

understand and keep in memory.

Figure 4.10 shows the user graphical user interface designed and implemented

for this experiment. The 24-channel audio ﬁles were produced beforehand (oﬀ-

line), being calibrated to the speciﬁc setup. Along with audio presentation,

the software also sent a series of triggers in synch with presentation timings

to the EEG software (Actiview, BioSemi) to mark the EEG measurement

Chapter 4. Methods 112

appropriately for the companion analysis.

Figure 4.10: Graphic User Interface used to acquire the data from participants.

Words are state buttons that alternates between green and red being saved as 1 or 0

respectively.

4.2.6 Questionnaire

At the end of each block (SNR × room condition) a three-item questionnaire

was presented to the participants; the English translation is shown in Ta-

ble 4.1. The questionnaire was translated from Zekveld and Kramer [318] to

Danish. The response to each question had a scale of 0 to 100 in integer units

Appendix F. The ﬁrst question was aimed to measure participants’ estima-

tion of their performance, referred to as “Subjective intelligibility” for the rest

of the text. The second question was to measure participants’ perception of

eﬀort, referred to as “Subjective eﬀort”. The third question provided to mea-

sure how often participants gave up during the test, referred to as “Subjective

disengagement”.

Chapter 4. Results 113

Table 4.1: The questionnaire for subjective ratings of performance, eﬀort and en-

gagement (English translation from Danish)

Question 1 How many words do you think that you understood correctly?

Question 2 How much eﬀort did you spend when listening to the sentences?

Question 3 How often did you give up trying to perceive the sentences?

4.2.7 Statistics

A linear mixed model [171, 233] (LMM) was used to investigate SNR and

RT eﬀects on performance and questionnaire. The eﬀects on diﬀerent alpha

bands through EEG power by SNR and RT were also explored through LMM

in the collaborative analysis performed by Seiﬁ Ala. SNR and RT were ﬁxed

factors, while participants were random factors in the model. Implemented in

MATLAB, the syntax for LMM was Dependent ∼ 1+SNR*RT+(1—Subject

ID), with Dependent being either performance or questionnaire. Both the SNR

(-5, 0, 5) and RT (-0.53, -0.03, 0.56) levels were re-centered around zero for

the model.

4.3 Results

This section highlights the ﬁndings concerning the study’s questions, the fea-

sibility of having a hearing in noise test virtualized in ﬁrst-order Ambisonics,

and the inﬂuence of degradation through SNR and Reverberation in the Speech

Intelligibility.

The participant’s behavioral performance (i.e., speech recognition accuracy)

demonstrated signiﬁcant eﬀects of SNR (β = 5.98, SE = 0.30, t

158

= 19.67, p <

0.001), and RT (β = −31.17, = 1.78, t

158

= −17.49, p < 0.001) and a signiﬁcant

interaction between the two (β = 1.76, SE = 0.43, t

158

= 4.04, p < 0.001).

Figure 4.11 presents the mean performance (percent correctly recalled words)

as a function of SNR for each room. Less signal degradation, whether higher

Chapter 4. Results 114

SNR or lower RT led to higher performance accuracy.

Figure 4.11: Performance accuracy based on percentage of correctly recalled words

as a function of SNR and RT (line color/shading). Error bars represent the standard

error of the mean. Lines/symbols are staggered for legibility and do not indicate

variation in SNR.

The statistical analysis of the results for subjective intelligibility (Figure 4.12),

subjective eﬀort (Figure 4.13), and subjective disengagement (Figure 4.14) are

shown in Table 4.2. All the measures show a signiﬁcant interaction between

SNR and RT. Lower signal degradation (higher SNR and lower RT) led to

higher subjective estimation of intelligibility performance accuracy, decreased

reported eﬀort and disengagement.

Table 4.2: Results of linear mixed model based on SNR and RT predictors estimates

of the questionnaire.

DF = 158 Self-report scales

Question

Predictor

Subjective intelligibility Subjective eﬀort Subjective disengagement

SNR

β = 5.71

SE = 0.42

t = 13.48

p <0.001

β = -5.60

SE = 0.41

t = -13.57

p <0.001

β = -5.78

SE = 0.48

t = -11.85

p <0.001

β = -33.74

SE = 2.47

t = -13.61

p <0.001

β = 23.58

SE = 2.41

t = 9.76

p <0.001

β = 33.39

SE = 2.85

t = 11.68

p <0.001

SNR x RT

β = 1.56

SE = 0.60

t = 2.57

p = 0.010

β = 1.50

SE = 0.59

t = 2.54

p = 0.012

β = -2.06

SE = 0.69

t = -2.94

p = 0.003

Chapter 4. Results 115

Figure 4.12: Subjective intelligibility as a function of SNR and RT (line

color/shading). Error bars represent the standard error of the mean. Lines/symbols

are staggered for legibility and do not indicate variation in SNR.

The subjective impression of how much eﬀort was required and how willing

they were to give up in each situation are presented in Figures 4.13 and 4.14,

respectively.

Figure 4.13: Subjective eﬀort as

a function of SNR and RT (line

color/shading). Error bars repre-

sent the standard error of the mean.

Lines/symbols are staggered for legibility

and do not indicate variation in SNR.

Figure 4.14: Subjective disengage-

ment as a function of SNR and RT

(line color/shading). Error bars rep-

resent the standard error of the mean.

Lines/symbols are staggered for legibility

and do not indicate variation in SNR.

Chapter 4. Discussion 116

The results show the statistically signiﬁcant contributions of reverberation and

SNR to perceived performance, eﬀort and disengagement. From Figures 4.12,

4.13, and 4.14, the self-report scales varied near-linearly with the signal degra-

dations across conditions, agreeing generally with the behavioral data (See

Figure 4.11).

The subjective eﬀort is related to the inverse of the reverberation time; the

more time the energy needs to dissipate in the environment, the greater the

perceived eﬀort. The results from all the self-report scale questions were highly

correlated to performance. Pearson skipped correlations [308] revealed a sig-

niﬁcant ρ coeﬃcient (See Table 4.3):

Table 4.3: Pearson skipped correlations between performance and self-reported ques-

tions.

performance vs

subjective intelligibility

performance vs

subjective eﬀort

performance vs

subjective disengagement

r 0.95 -0.79 -0.94

CI 0.93, 0.96 -0.84, -0.74 -0.96, -0.92

4.4 Discussion

This study presented an interesting challenge to the researchers. The pilot

data pointed to the direction of the virtualization not rendering the correct

sound, especially not the correct sound pressure level. The setup was retested

and investigated in diﬀerent positions, and the problem was identiﬁed. The

ﬁrst-order Ambisonics rendering has a relatively small sweet spot. Thus par-

ticipants were monitored to be in the correct position during the testing. The

sweet spot capabilities in terms of the correct overall SPL reproduction pre-

sented limitations of plus-minus 1 dB up to 5 cm and plus-minus 3 dB to the

target SPL up to 10 cm oﬀ center. Although not testing the exact Ambisonics

implementation through a diﬀerent performance measure, the ﬁndings agree

with literature observing contrasts caused by the reproduction method in sim-

Chapter 4. Discussion 117

ilar distances out of the center. As a reference, Grimm et al. [97] analyzed

simulated Ambisonics environments with diﬀerent numbers of loudspeakers,

studying its inﬂuence on a representative hearing aid algorithm. It showed a

decrease in SNR errors when increasing loudspeakers and decreasing frequency.

A bandwidth of 2 kHz in the central listening position, 12 loudspeakers would

be required for HOA. If 24 loudspeakers are available, the bandwidth in the

central listening position would be 6 kHz. Laurent et al. [276] analyzed the

reconstruction error to assess the rendering system’s frequency capabilities.

A KEMAR was ﬁtted with a hearing aid, without processing, to collect the

impulse responses. Regarding range, a third-order implementation with 29

loudspeakers decreased from 3,150 Hz in the center to 2,500 Hz when posi-

tioned 10 cm from the center.

Tests that involve separated sound sources and are auralized and virtualized by

loudspeaker setups need to be veriﬁed in terms of sweet spot size to the speciﬁc

sound parameters (e.g., RT and SPL). An oﬀ-centered or moving head can, in

an Ambisonics 1st order auralization, easily encounter a spot in space where,

for example, the wave ﬁeld combination may partially cancels one or more

maskers increasing the SNR, even if the intended SNR is low (See Figure 4.7).

In other oﬀ-center spot it could also be possible to partially cancel the target.

These distortions could profoundly impact the results, and not represent what

would be achieved in the real scenario that was being simulated.

For normal hearing participants, a more psychologically oriented psychoacous-

tic auralization method such as lower order Ambisonics can provide the desired

acoustic impression insofar as objective and subjective performance when the

calibration is performed, and the setup limitations (e.g., very restricted sweet

spot) are respected. An investigation of performance in oﬀ-center positions

using hearing impaired participants would be an important next step towards

understanding a broad clinical application of this method.

Chapter 4. Concluding Remarks 118

Participants were tested in three diﬀerent SNRs (-8, -3, +2 dB) and three vir-

tual rooms (with RTs of 0, 0.5, and 1.1 s). The more the manipulated signal

was degraded (lower SNR and higher RT), the more demanding the listening

conditions became, which led to lower the participant’s speech intelligibility.

A questionnaire was used as a subjective measure of eﬀort. Comprehensively,

participants reported increased speech intelligibility, less cognitive eﬀort, and

less tendency to disengagement when diminish the signal degradation. That

denotes that if they could recall the speech well, they perceived that they

performed well and also spent less eﬀort. The results from all three ques-

tions within the questionnaire were strongly correlated (either positively or

negatively) to the speech intelligibility of the participants. They signiﬁcantly

changed with both SNR and RT and the interaction between them. When

asked about subjective impressions of each block, the participants demon-

strated to have perceived the proposed signal degradation both in SNR and

RT. That is in line with the studies from Zekveld et al. [319], Holube et al. [121],

Neuman et al. [203], Kwak et al. [149], Nicola & Chiara’s study [204] and Picou

& Ricketts [229]. Furthermore, studies that cross objective measurements of

physiological parameters in the literature associated with changes in eﬀort can

have divergent outcomes, as discussed in Chapter 2. From that discussion, it is

speculated that these diﬀerent methods, proposed to achieve a proxy to listen-

ing eﬀort, are sensitive separated aspects of a complex global process [12, 224].

Another explanation would be the minimization of eﬀort utilized by the par-

ticipant through the heuristic strategies in the subjective method [192], and

lastly, the eﬀect of working memory being related diﬀerently to diﬀerent meth-

ods [53, 186]. A separated study by Tirdad Seiﬁ-Ala from this combined ex-

periment examined the correlation between objective (physiological responses

of the brain) and subjective paradigms.

Chapter 4. Concluding Remarks 119

4.5 Concluding Remarks

In this study, nine levels of degradation were imposed on speech signals over

speech maskers separated in space and virtualized. Three diﬀerent SNRs (-8,

-3, +2 dB) and three diﬀerent simulated rooms (with RTs of 0, 0.5, 1.1 s)

were used to manipulate task demand. The speech intelligibility was assessed

through a word-scored speech-in-noise test performed in a 24 loudspeaker setup

utilizing Ambisonics ﬁrst-order. The results showed a high correlation between

participants’ performance and responses to questions about subjective intel-

ligibility, eﬀort and disengagement. The main eﬀects and interaction of SNR

and RT were demonstrated on all questions. Furthermore, it was observed

that the reverberation time inside a room impacts both speech intelligibility

and listening eﬀort. This study demonstrated the possibility of virtualizing a

combination of sound sources in low order Ambisonics and extracting quality

behavioral data.

Chapter 5

Iceberg: A Hybrid Auralization

Method Focused on Compact

Setups.

5.1 Introduction

People usually wear their hearing devices in spaces very diﬀerent from the

laboratories’ soundproof booths in everyday life. Additionally, the everyday

sounds are more complex and diﬀerent from the pure tones, words, and phrases

without context utilized in many hearing tests. Therefore, hearing research has

increasingly aimed to include acoustic verisimilitude on auditory tests to make

them more realistic and/or ecologically valid [61, 79, 101, 177, 212, 217]. Thus,

they can evaluate new features and algorithms implemented on hearing devices

and experiment with diﬀerent ﬁttings and treatments while maintaining their

repeatability and control.

One can utilize a particular auralization technique to create reproducible sound

120

Chapter 5. Introduction 121

ﬁles in a listening area. These sounds attempt to mimic the acoustical charac-

teristics of environments (from actual recordings or acoustic simulations). It

can then be played through a set of loudspeakers or pair of headphones, cre-

ating both the subjective impression and objective representation of listening

to the intended sound environment [293].

Through an auralization method, it is possible to create a sound ﬁle containing

spatial information about the scene and a series of details about the conﬁgura-

tion of the reproduction system [293]. The reproduction system includes, for

example, the number of loudspeakers and their physical position, the number

of audio channels available, and the distance from the loudspeakers to the lis-

tening position. The size of the eﬀective listening reproduction area - where

the auditory spatial cues of the scene are most accurate - is usually called

the ”sweet spot” [253]. The spatialization accuracy is diﬀerently aﬀected by

diﬀerent systems as well as auralization methods [65, 97, 166, 275, 276].

The auralization method can be decisive in the reproduction system choice; for

example, certain methods require certain numbers of loudspeakers [62, 217].

Consequently, the auralization method can be a limiting factor depending on

the tests or experiments. A dedicated setup capable of handling diﬀerent

auralization methods with a large listening area [188] may require an excessive

amount of funding and physical space. These requirements can be a limiting

factor to conducting research and developing innovative treatments.

This chapter proposes a compact setup with a hybrid auralization method. It

is characterized in some conditions (RTs, presence of a second listener, and

listener position) by considering the intended use in auditory evaluations as in

the previous chapter. The setup aims to reproduce sound scenes maintaining

spatial localization and creating an immersive sound environment from either

a scenario in an actual room or virtual rooms created in acoustic software.

Chapter 5. Iceberg an Hybrid Auralization Method 122

5.2 Iceberg an Hybrid Auralization Method

The Iceberg auralization method combines two well-known methods: VBAP

and Ambisonics. In Chapter 3, VBAP and Ambisonics binaural cues were

objectively evaluated. The VBAP method was found to render accurate cues

in the center position, even with a second listener inside the array. That

corroborates the use of VBAP to increase tests’ ecological validity in auditory

tests [134]. On the other hand, Ambisonics delivered less precise localization

cues, imposing more restrictions on the listener’s position. The results are

in line with literature presenting poor localization but high immersiveness

from low-order Ambisonics [104, 105] and, conversely, lesser immersiveness

and greater localization accuracy from VBAP [89, 104]. Therefore, the idea

here is to provide an auralization that contains temporal and spectral features

of the sounds encoded through VBAP while the spaciousness provided through

the reverberation envelope is encoded through Ambisonics.

This speciﬁc combination of auralization methods has also been considered to

decrease the number of necessary loudspeakers for a setup that requires regular

hearing devices. At the same time, the setup may allow some degree of head

movement without the need for tracking equipment. That is a countermeasure

to overcome common limitations in ordinary auditory test spaces [316].

5.2.1 Motivation

The primary motivation for creating this auralization method was to test hear-

ing aid users in typical situations while wearing hearing devices in a small

setup. Therefore, the method is loudspeaker-based, but at the same time,

the number of loudspeakers and the system complexity were also constraints.

The theoretical support for combining these auralization methods and propos-

Chapter 5. Iceberg an Hybrid Auralization Method 123

ing the smaller virtualization setup is gathered from room acoustic parameters

and psychoacoustics principles presented in the review and during this chapter.

These parameters and principles led to a system able to use RIRs from simu-

lated environments (spaces that may only exist in a computer) and recorded

RIRs from real ones. The initial Iceberg focus is on tests that manipulate

sound scenarios to evaluate speech intelligibility masked by noise from static

positions as tested with low-order Ambisonics in Chapter 4.

5.2.2 Method

The Iceberg method is a relatively easy-to-use algorithm that can be intro-

duced to test environments with a simple calibration process. The virtual-

ization system presented auralized ﬁles in a quadraphonic array with loud-

speakers positioned at 0, 90, 180 & 270

◦

(see Figure 5.1). Other horizontal

setup arrangements can be implemented depending on the need, considering

the system’s angle rotation, frequency response, and the potential variation

in localization accuracy. Although there is a minimum number of necessary

loudspeakers (four), the method can be used to auralize ﬁles to setups with an

extended loudspeaker number. The presented algorithm was implemented in

MATLAB (Mathworks).

Figure 5.1: Top view. Loudspeakers position

on horizontal plane to virtualization with pro-

posed Iceberg method.

The proposed loudspeaker

setup had a radius of 1.35

m. Other distances need to

be evaluated regarding the

system frequency response.

The proposed Iceberg’s implementation derives an appropriate multi-channel

audio signal with speciﬁc information from a sound and its reﬂections (inci-

Chapter 5. Iceberg an Hybrid Auralization Method 124

dence angle, sound energy, spatial and temporal distribution). These param-

eters can be encoded into a sound ﬁle with the reproduction setup’s speciﬁc

calibration values and positioning orientation. Finally, the auralized ﬁle can

be reproduced (virtualized) as spatial sound.

5.2.2.1 Components

The Iceberg method proposed is a hybrid auralization method, a combination

of VBAP and ﬁrst-order Ambisonics; section 2.3.2.2 reviews the derivation

of both methods. Both techniques are based on the panorama of amplitude.

The main diﬀerence is in the mathematical formulation of the gains applied

to the amplitude of each sound source. VBAP treats the reproduced sound

as a unitary vector in a two- or three-dimensional plane (Equations 2.4 and

2.7, respectively). The weights applied to the amplitude of the signal at each

loudspeaker are derived from the tangent law. It is traced as a vector from

the nearest available sources between the listening position and the desired

source position (Equation 2.3). On the other hand, Ambisonics utilizes all

loudspeakers available to compose the sound ﬁeld. The method combines the

amplitude of the sources, calculating their weights according to the sum of

spherical harmonics (Equation 2.9) that represents the pressure ﬁeld formed by

the sound wave (Equation 2.8). While VBAP concentrates the energy between

two loudspeakers in its 2D implementation, Ambisonics spreads it through

all available loudspeakers. That leads to a more immersive experience on

Ambisonics, while the VBAP can better represent the sound source direction.

5.2.2.2 Energy Balance

The energy balance between the methods is calculated based on the Ambisonics

ﬁrst-order impulse response (See example in Figure 5.2); on the left is the

Chapter 5. Iceberg an Hybrid Auralization Method 125

impulse response (or decay curve) and is not the decay of the squared value

of the sound pressure signal. On the right, the 10log curves (h

(t)) for the

diﬀerent channels. Note that in these curves, the maximum level is 0 [dB], as

the interest is in the time it takes for the power to drop by 60 [dB]. Also, note

that there is a small time gap between time 0 [s] and when the energy value of

h(t) is maximum. This interval corresponds to the time it takes the sound wave

to travel between the source and the receiver and allows an estimate of the

distance between them. From recorded IRs, the gap also includes the system

delay, which should be compensated. That was the choice since the impulse

response of an environment can easily be acquired utilizing an Ambisonics

ﬁrst-order microphone array. Furthermore, it is possible to ﬁnd commercially

available acoustic software tools to simulate sound environments capable of

exporting impulse responses in Ambisonics format.

Figure 5.2: Normalized Ambisonics ﬁrst-order RIR generated via ODEON soft-

ware. Left panel depicts the waveform; right panel depicts the waveform in dB.

The system’s design requires an RIR to be split into two parts. The ﬁrst part

contains the amount of energy to be delivered through VBAP. The second

part will be computed through Ambisonics. From the reﬂectogram, the time

representation of the latency and attenuation of the direct sound (DS), early

reﬂections (ER), and late reﬂections (LR; see Figure 5.3), it is possible to ﬁnd

Chapter 5. Iceberg an Hybrid Auralization Method 126

the point in time representing the direct sound (the ﬁrst peak) and then sep-

arate it correctly from the rest of the RIR. Although splitting the RIR into

DS and remainder may be the most straightforward method, the achieved re-

sults were initially perceived in personal experience as unnatural highlighted

“dry” (not reverberant) sound from a deﬁned position followed by really dis-

tant/disconnected reverberation, counter to the aims of a more ecologically

valid sound reproduction. Thus, in the proposed method the ER part was

included with the DS part.

Figure 5.3: Reﬂectogram split into Direct Sound Early and Late Reﬂections.

The late reﬂections of an RIR refer to the signal wavefronts reﬂected and scat-

tered several times across the diﬀerent possible paths. These reﬂections overlap

each other, and as time progresses, successive wavefronts interact with any sur-

face, increasing reﬂection order, changing direction and decreasing remaining

sound energy.

The literature indicates that a psychoacoustical approximation of the time

point in a speciﬁc RIR when the human auditory system can no longer distin-

guish single reﬂections due to reﬂection density [38]. Lindau [156] proposed a

transition point in time (transition time (t

)) based on mean free path length

for the wavefront (Equation 5.1).

20V

+ 12 [ms] , (5.1)

Chapter 5. Iceberg an Hybrid Auralization Method 127

where, V is the volume of the room in m

, and S is the surface area inside the

room in m

The minimum necessary order of reﬂections to represent a uniform and isotropic

sound ﬁeld that leads to diﬀuse reverberation from an Image Source (IS) model

is 3. That agrees with observations from Kuttruf [148] on the specular reﬂec-

tions’ contribution to diﬀuse energy in an RIR.

This approach was implemented in a similar hybrid method by Pelzer et

al. [221]. Another method developed by Favrot [79] also uses the IS order

information from simulated RIR computed with ODEON software. Its IS re-

ﬂection order information provides a point to obtain a segment of the ﬁle with

the late reﬂections envelope used by the system to deliver a hybrid multi-

channel RIR. These methods consider RIR and mix speciﬁc stimuli to the

output, as does the proposed method. Other hybrid auralization methods

such as DirAC [243] consider the recording of a sound event (in Ambisonics)

and drive the reproduction based on energy analysis spanning all sound source

directions. Thus, DirAC is intended to primarily work with recorded scenes

instead of convolutions with RIR.

5.2.2.3 Iceberg proposition

The proposed Iceberg method, however, uses neither the t

method, which is

dependent on the volume of the rooms and the IS simulated reﬂection order,

nor the LR envelope time, derived from an IS simulation. Instead, a diﬀerent

parameter is proposed that allows generalizing to both recorded and simulated

Ambisonics RIRs.

Parameters of clarity and deﬁnition are metrics to determine early/late energy

balance [43]. However, the ﬁxed time of 50 or 80 milliseconds is not appropriate

Chapter 5. Iceberg an Hybrid Auralization Method 128

to represent the transition point (from early to late reﬂections) on every RIR,

as the slope will diﬀer and depend on many factors [45]. The transition point

changes as the amount of energy and the decay distribution change from RIR to

RIR. A similar parameter that is not time ﬁxed is the T) given by Equation 2.15

(see Section 2.3.3). This parameter is also derived from the squared RIR by

calculating the transition point from early to late reﬂections represented as

the RIR’s center of gravity. Therefore, the method’s name is given because

of the singularity of the RIRs. They present a center of gravity on its power

decay representation, which is similar to the physical blocks of frozen water

called icebergs. The center of gravity is the equilibrium point between the

gravity force and the water buoyancy for icebergs [34]. This representation is

translated to the Iceberg method as the transition point between early and

late reﬂections from an RIR.

This process entails an RIR applied through multiplication in the frequency

domain, equivalent to a convolution in the time domain, to a sound that can

be virtualized through the system. The ﬁrst action of the method’s algorithm

is the identiﬁcation the center time Ts in the channel relative to the omnidi-

rectional channel of the Ambisonics RIR. A schematic overview of the method

is presented in Figure 5.4.

Chapter 5. Iceberg an Hybrid Auralization Method 129

Figure 5.4: Iceberg’s processing Block diagram. The Ambisonics RIR is treated,

split, and convolved to an input signal. A virtual auditory scene can be created by

playing the multi-channel output signal with the appropriate setup..

Figure 5.5 shows an example of the RIR relative to the omnidirectional in-

put channel simulated through ODEON V.12 [59] relative to the simulated

restaurant dining room used in Chapter 4 with 1.1 seconds of reverberation

time.

Figure 5.5: Omnidirectional channel of Ambisonics RIR for a simulated room. The

blue line indicates the part that previously selected the calculated Center Time, hence

indicated as the direct sound plus the early reﬂections. The orange line indicates the

late reverberation part of the RIR.

Chapter 5. Iceberg an Hybrid Auralization Method 130

Figure 5.6 presents an example of the Ambisonics RIR in the left column, the

omnidirectional channel relative to the DS+ER in the middle column. The

right column graphs represent the four channels late reﬂections part of the

Ambisonics RIR.

Figure 5.6: First column: four channels Ambisonics RIR. Middle column: omni-

directional channel (DS+ER part). Right column: four channels Ambisonics RIR

(LR part).

In the sequence, the method ﬁrst splits the RIR based on the TS. Then the

direct sound and the early reﬂections are convolved with the signal to be repro-

duced. In this step, only the omnidirectional channel is used. Finally, the sig-

nal is processed using VBAP to provide its directional properties. The VBAP

method utilized was implemented in [237]. The VBAP output is two-channel

panned audio that is sent to channels of the corresponding loudspeakers. The

output signal corresponds to the relative full scale of the panned signal if the

provided Ambisonics RIR is normalized, or the absolute value in case of an

un-normalized RIR. With the normalized RIRs, calibration of a sound pres-

sure level is required, and the reproduction level can be set accordingly to the

application needs. Assuming a coherent sum between two loudspeakers that

are set to reproduce the scaled signal to a predeﬁned level, a proportion is

computed as follow:

Chapter 5. Iceberg an Hybrid Auralization Method 131

LS1 = 20 log

(10

level/20

∗ sin

θ), (5.2a)

LS2 = 20 log

(10

level/20

∗ cos

θ), (5.2b)

where the user sets the level in dB SPL and θ is the incidence angle. A similar

level calibration recording a pure tone from a calibrator with a microphone

to ﬁnd a system’s α coeﬃcient (as explained in 3.2.3) will allow playing the

signal over each loudspeaker with the intended level. A frequency ﬁlter for each

loudspeaker is also possible if the loudspeakers’ FRF needs to be individually

adjusted to achieve a ﬂat(ter) response.

The second part of the impulse response is then convolved with the signal,

all four channels in the proposed quadraphonic system being used. First, an

Ambisonics decoder matrix observing the loudspeakers’ position is created.

Thus, the convolved signal is decoded from its bFormat to aFormat. The

implementation utilized in the algorithm to create the decoder matrix and to

decode the signal uses the functions from the Politis [237] work.

The separated signals are then merged being ready to be reproduced.

Figure 5.7 show an example for an auralization of ﬁve seconds of the Interna-

tional Speech Test Signal (ISTS) [120]. The top graph is the original signal,

and the mid-top is the signal convolved with the DS and ER of the omnidi-

rectional channel of the Ambisonics RIR. The envelope is minimally aﬀected

by the ER. The mid-bottom shows the signal convolved to the LR part of

four channels and decoded from Ambisonics bFormat. The diﬀuse nature of

the Ambisonics-generate LR evident in the smoother overall envelope. The

bottom graph shows the result of the Iceberg method, the merged signal.

This process provides an auralized ﬁle that should be reproduced through an

Chapter 5. Iceberg an Hybrid Auralization Method 132

Figure 5.7: Iceberg method example. Top graph: original signal. Mid top graph:

DS+ER part (VBAP). Mid bottom graph: LR part (Ambisonics). Bottom part

(Iceberg).

equalized and calibrated setup. An equalization and calibration proposal is

described in the Section 5.2.3 and can be applied to similar setups with equiv-

alent hardware. However, the results may vary depending on hardware quality,

loudspeaker ampliﬁcation, and frequency response. In this work, the electroa-

coustical requirements (7.2.2) and Reference listening room (8.2) from the

Recommendation ITU-R 1116-3 [126]: Methods for the subjective assessment

of small impairments in audio systems were observed. The frequency-speciﬁc

reverberation times were lower than the Recommendation: 0.04 s from 0.2-

4 kHz (0.08 s at 0.125 kHz) and 0.18 s in the Recommendation. The anechoic

characteristic of the room was intentionally chosen in this case to evaluate

reverberation in the virtualization setup. A setup within a diﬀerent space

will have diﬀerent room acoustics characteristics. The experimenter can com-

pensate for the need for greater reverberation by controlling the RIRs input.

The electroacoustical requirements for the loudspeakers are also relevant as

they aim to guarantee the correct frequency reproduction or the possibility

of compensating the frequency response with the appropriate hardware. The

room proportions are also essential when setting a test environment, espe-

cially if the reproduction will include low frequencies aﬀected by the room’s

Chapter 5. Iceberg an Hybrid Auralization Method 133

Eigentones (standing waves).

The address https://github.com/aguirreSL/HybridAuralization contains an ex-

ample and the necessary resources to auralize ﬁles according to this Iceberg

method. This study utilizedAmbisonics ﬁrst-order impulse responses generated

with the ODEON V.12 software. The choice was made by convenience, and it

can be extended to any equivalent Ambisonics RIR, simulated or recorded.

The resulting RIR from ODEON is normalized. With that, the user can play

a sound on a diﬀerent level (from the simulated one) without rerunning the

simulation using the normalized version. As an option, the method can denor-

malize it (dividing the RIR by its corresponding factor provided in ODEON

grid [159].). The denormalized result sound will be auralized to the level sim-

ulated in ODEON (or equivalent software).

5.2.3 Setup Equalization & Calibration:

The setup can include a calibration and equalization procedure that is included

in the MATLAB scripts to ensure a correct sound level reproduction and also a

ﬂatter frequency response from the system’s loudspeakers, avoiding additional

undesired coloration artifacts. First it was calculated a factor to transform the

acquired signals from full scale to dB SPL. This step consists in recording a

pure tone at a speciﬁc frequency (1 kHz) with a known input level of 1 Pa,

and calculating a factor to convert the input from Full-Scale to Pa. The term

indirect refers to the fact that this calculated factor is applied to all frequencies,

under the assumption that the setup (microphone, pre-ampliﬁer, power supply,

and AD/DA converter) has a ﬂat frequency response in the audible frequency

range. To calculate the conversion factor a sound pressure calibrator device

(in this case the B&K 4231) was connected to the microphone (1/2” B&K

Chapter 5. Iceberg an Hybrid Auralization Method 134

4192 pressure-ﬁeld and a pre-ampliﬁer type 2669, supplied by power module

B&K 5935). That provided a 93.98 dB SPL signal, which corresponds to 1 Pa.

The calibration factor (α

rms

) was calculated as in Equation 5.3. Although this

step was not needed to the frequency equalization, it was convenient as once

measured all the following measurements were performed without the need of

entering the room.

rms

RMS(v(t)

1kHz

)





, (5.3)

The next step consists in equalize the frequency response of each loudspeaker.

A RIR from each loudspeaker was measured and based on that an inverted

FIR ﬁlter was individually created to be applied to the signals that will be

reproduced. The frequency response was converted to its third-octave version,

normalized, and inverted to create a vector with 27 values from 50 Hz to 20

kHz. These vectors contained the correction values in the frequency domain

and can be applied to any input signal. To apply this corrections a Piecewise

Cubic Hermite Interpolating Polynomial (PCHIP) was used in MATLAB to ﬁt

the values to the given input. Figure 5.8 presents an example of the normalized

third octave moving average RIR acquired with a loudspeaker (blue line), the

same RIR but acquired with a signal that was ﬁltered (red line), and the ﬁlter

frequency values obtained with the inversion of the original RIR (black line).

Chapter 5. Iceberg an Hybrid Auralization Method 135

Figure 5.8: Loudspeakers normalized frequency response and inverted ﬁlter. Doted

lines represent ITU-R 1116-3 limits.

Figures 5.9 and 5.10 shows the moving average of each loudspeaker’s normal-

ized frequency response without and with the ﬁlter, respectively.

Figure 5.9: Loudspeakers normalized frequency response (colored solid lines), doted

lines represent ITU-R 1116-3 limits.

Chapter 5. Iceberg an Hybrid Auralization Method 136

Figure 5.10: Loudspeakers normalized frequency response with frequency ﬁlter cor-

rection (colored solid lines), doted lines represent ITU-R 1116-3 limits.

As the ampliﬁcation to each active loudspeaker is individually controlled it is

possible that a same ﬁle could be reproduced at a diﬀerent sound pressure level

(if someone inadvertently or accidentally change the volume control button

directly in the loudspeaker for example). Since the α

rms

was already calculated

and it was possible to convert a signal from FS to Pa, and consequently, to dB

SPL, and vice-versa the individual loudspeakers’ SPL were measured with a

signal deﬁned to be played at 70 dB SPL Equation 5.4.

signal(t) =



signal(t)

rms(signal(t))

70−dBperV

20µ



(5.4)

where

is the level factor to the loudspeaker l with initial value = 1;

dBperV = 20 log



rms

20µ



The signal(t) was played through a loudspeaker l and simultaneously recorded

with the microphone S

(t); the SPL of the recorded signal was calculated as

follows

Chapter 5. Iceberg an Hybrid Auralization Method 137

SPL

= 20 log

(t)[FS]α

rms





20[µPa]

[dB] , (5.5)

Ten measurements were sequentially performed with each loudspeaker at inter-

vals of 1 s; another iteration of measurements were performed if the measured

SPL exceeded the tolerance of 0.5 dB on any of the measurements. A step

of ± 0.1 [FS] is set to update Γ

in its next iteration according to the SPL

obtained.

Chapter 5. System Characterization 138

5.3 System Characterization

The Iceberg auralization method in a four-loudspeaker system (minimum re-

quired) was evaluated for its capabilities to reproduce the intended reverbera-

tion time and the appropriate binaural cues. This section describes the system

setup, and the conditions experimented with utilizing the Iceberg method. The

method’s accuracy at the optimal and sub-optimal positions was considered

in this characterization as well the impact of the RT. Furthermore, placing

a second listener inside the ring was investigated to support a more ecolog-

ical situation. By the end, a complementary study for those conditions was

conducted with an aided mannequin to supplement the objective data as the

pandemic prevented subjective data collection.

The present study used the ITA-Toolbox [29] for signal acquisition and process-

ing. To further enhance the accuracy of the localization estimates, a MATLAB

implementation of the May and Kohlrausch [182] localization model from the

Auditory Modeling Toolbox (AMT, https://www.amtoolbox.org) [287]) was

also employed. The May model is speciﬁcally designed to be robust against

the detrimental eﬀects of reverberation on localization performance, making it

an ideal choice for supplementing the objective data gathered in the present

study. The reverberation, or the persistence of sound after its initial source

has ceased, was a parameter in this test that could signiﬁcantly distort the

estimated location of a sound source. The May model accounts for reverbera-

tion’s inﬂuence through frequency-dependent time delay parameters, enabling

more accurate localization estimates in reverberant environments. By incorpo-

rating the model in our analysis, we supplemented the objective data gathered

through signal processing with an additional layer of modeling that allowed a

relative comparison with previous studies.

The main objective of an auralization method and its virtualization setup is

Chapter 5. System Characterization 139

to deliver appropriate spatial awareness to human listeners. The natural step

for this would be to verify and validate the method. Unfortunately, special

conditions were in place during the course of this study; due to COVID-19 re-

strictions, validation tests with participants were not feasible. The Section 5.5

extends the system veriﬁcation and analysis to a targeted application of hearing

aid research. Although it does not replace a subjective impression validation

and analysis, it can help understand and predict the system’s behavior in a

typical use case for hearing research, which is the user with hearing aids.

5.3.1 Experimental Setup

The proposed method was implemented, and the tests were conducted at Erik-

sholm Research Centre in Denmark. The test environment was an anechoic

room (IAC Acoustics) with inner dimensions of 4.3 m × 3.4 m × 2.7 m. Sig-

nals were routed through a sound card (MOTU PCIe-424) with a Firewire 440

connection to the MOTU Audio 24 I/O interface and played via loudspeak-

ers Genelec model 8030C (Genelec Oy, Iisalmi, Finland). The well-controlled

sound environment was appropriate for the assessment of small impairments in

audio systems, although the acoustic properties of the room exceed the sound

booths and rooms commonly encountered in audiology clinics [316].

5.3.2 Virtualized RIRs & BRIRs

A set of 72 room impulse responses and 72 binaural room impulse responses

were acquired through the system separated over 5 degrees angles around the

center position assuming x as lateral axis and y the front-back (mid-saggital)

axis of a person inside the ring. Moreover, the same number of RIRs and

BRIRs were measured at oﬀ-center positions.

Chapter 5. System Characterization 140

The virtualized RIRs and BRIRs were acquired utilizing a logarithimc sweep

signal (50-20000 Hz, 2.73 s (FFT Degree 18, Sample Frequency 96 kHz)) [194]

as input. The signal was auralized to each angle in the Iceberg method for

the same three spaces as in Chapter 4: a classroom (9.46 m x 6.69 m x 3.00

m) with an overall Reverberation Time RT of 0.5 s, a restaurant dining area

(12.19 m x 7.71 m x 2.80 m) with an overall RT of 1.1 s, and an anechoic

room (4.3 m x 3.4 m x 2.7 m) with an ideal overall RT of 0.0 s. All rooms

were acoustically simulated in ODEON software V.12, that generated the am-

bisonics RIRs representing each mentioned source-receptor conﬁguration. The

absorption coeﬃcients of the room surfaces are listed in Appendix E.

The initial step to acquiring the RIR and BRIR was to auralize the sweep

ﬁle utilizing the Iceberg method to the desired positions (72 angles around

the center) in three diﬀerent room conditions. Then play it through the four

loudspeakers positioned in the front (0°), left (90°), back (180°) right (270°)

counter-clockwise angles. The auralized version of the sweep should correspond

to the signal played in the virtual environment as the reverberation added by

the anechoic room is negligible. After that, the recorded ﬁle was de-convolved

with a zero-padded version of the raw sweep (See Figure 5.11).

The playback and recording utilized the maximum sampling rate supported

on the AD/DA system (96,000 Hz) as the diﬀerence in time is in the µs scale.

Therefore, the step size in time provided in microseconds is given by step size =

(1/96, 000) ∗ 1, 000, 000 = 10.42 µs. The created sweep duration was 2.731 s

(FFT Degree = 18).

Chapter 5. System Characterization 141

Figure 5.11: BRIR/RIR acquisition ﬂowchart: Iceberg auralization method.

A manikin with artiﬁcial pinnae (HATS model 4128-C; Br¨uel and Kjær) was

used to record the binaural ﬁles. Also, a second listener was simulated dur-

ing the tests with a diﬀerent manikin (KEMAR; GRAS), (See Figure 5.12).

The HATS recordings were calibrated as described in Section 3.2.3 following

Equations 3.1a and 3.1b.

Figure 5.12: BRIR measurement setup: B&K HATS and KEMAR positioned in-

side the anechoic room.

Chapter 5. System Characterization 142

5.3.3 Conditions

The auralized ﬁles were then recorded under the following conditions:

• Optimal position (alone and centered)

• Optimal position (centered) accompanied by a second listener

• Oﬀ center positions alone

The positions grid can be visualized in Figure 5.13.

Figure 5.13: Measurement positions: Obtained through virtualized sound sources

with Iceberg method (VBAP and Ambisonics) in a four-loudspeaker setup.

The most accurate performance is theoretically expected from optimal posi-

tion. These techniques provide virtualization assuming the receptor (listener)

is in the center of the loudspeaker ring [65, 241]. Adding a second listener

into the reproduction area and/or moving the primary listener away from the

center can challenge the system’s ability to render the scene as intended. The

following sections presents and discusses the system’s capabilities to repro-

duce Iceberg auralized ﬁles by measuring the binaural cues and RT in diﬀerent

conditions.

Chapter 5. System Characterization 143

5.3.4 Reverberation Time

A room’s characteristic wave ﬁeld pattern can aﬀect the human perception of

a reproduced sound. Room acoustics can alter attributes related to spatial

perception. For example, a recorded sound has almost no chance of being cor-

rectly reproduced if the reproduction room has stronger reverberation than the

recorded one. Also, reverberation overshoot can smear the perceived direction

of a sound source, as early reﬂections would be heightened in this case [242].

The RT was calculated from impulse responses measured within the three vir-

tualized environments (note that the simulated environments were aimed to

present RT of 0, 0.5, and 1.1 seconds). Reverberation time was calculated

using the ITA Toolbox. The parameters were set as follows: Frequency from

125 Hz to 16 kHz, one band per octave, and 20 dB threshold below maximum.

The reverberation time was shown to be stable in this virtualization setup. An

approx. 0.08 s Overall RT can be observed for the anechoic simulation (0 s RT).

That is most likely driven by the presence of the hardware inside the anechoic

room: loudspeakers and wood base for the chair, although covered with foam.

The overall reverberation time was measured without an omnidirectional sound

source. To circumvent this limitation, the measurement was repeated utilizing

all 24 loudspeakers as sound source, one at a time. The overall RT in this case

was considered as the maximum value across frequencies in octave bands from

125 Hz to 16k Hz. Figure 5.14 presents the boxplot of the measured values

in relation to its position inside the room. Rows represent the aimed RT (0,

0.5 and 1.1 s). The top line presents results without lateral displacement, the

middle line presents the results according to a lateral displacement of 2.5 cm

from the center and the bottom line presents the results according to a lateral

displacement of 5 cm from the center.

Chapter 5. System Characterization 144

Figure 5.14: Reverberation Time environments measured with ﬁles produced with

Iceberg method and virtualized in four-loudspeakers.

Table 5.1 present the median of the values of the overall RTs. Therefore,

it is possible to notice that virtualized environment RTs’, tend to be stable,

and to the measured conditions, under the just noticeable diﬀerence JND of

5% [264, 265] across positions inside the room.

Table 5.1: Reverberation Time in three virtualized environments in diﬀerent posi-

tions inside the loudspeaker ring.

RT = 0 RT = 0.5 RT = 1.1

Position [cm] Overall RT [s]

x=0.0; y=0.0 0.085 0.519 1.114

x=0.0; y=2.5 0.085 0.519 1.111

x=0.0; y=5.0 0.085 0.526 1.113

x=0.0; y=10.0 0.084 0.526 1.147

x=2.5; y=0.0 0.085 0.531 1.120

x=2.5; y=2.5 0.086 0.529 1.114

x=2.5; y=5.0 0.084 0.559 1.148

x=2.5; y=10.0 0.083 0.546 1.157

x=5.0; y=0.0 0.085 0.537 1.139

x=5.0; y=2.5 0.085 0.538 1.138

x=5.0; y=5.0 0.085 0.548 1.138

x=5.0; y=10.0 0.084 0.552 1.147

Chapter 5. Main Results 145

5.4 Main Results

This section presents the results based on the mannequin positions (center

and oﬀ-center) and conditions (HATS and HATS with KEMAR) to angles

referenced clockwise.

5.4.1 Centered Position

5.4.1.1 Interaural Time Diﬀerence

The blue line in Figure 5.15 represents the result of the Interaural Time Diﬀer-

ence ITD ﬁltered with a 1 kHz low-pass ﬁlter virtualized through the proposed

system.

Figure 5.15: Interaural Time Diﬀerence under 1 kHz as a function of azimuth

angle for a HATS Br¨uel and Kjær TYPE 4128-C in the horizontal plane through a

proposed Iceberg auralization method on a 4 loudspeakers setup. The red line is the

ITD results with real loudspeakers (without virtualization). According to the sample

rate, the blue and red shaded areas in are the conﬁdence intervals. The black line

represents the analytical ITD values.

Chapter 5. Main Results 146

Wang and Brown [297] deﬁned the analytical ITD (black line in ﬁgure) (See

Equation 5.6) considering a centered, perfect sphere of 10.5 cm of radius (a)

and sound propagation velocity (c) of 340 m/s, θ is the angle in radians.

ITD =





2 sin(θ) (5.6)

The maximum absolute diﬀerence found is 170 µs, representing a mismatch

around 15º on the given angle. The calculated average diﬀerence is 67 µs,

representing a diﬀerence of around 7º in localization.

Figure 5.16: Interaural Time Diﬀerence at 1 kHz as a function of azimuth angle

for a HATS Br¨uel and Kjær TYPE 4128-C in the horizontal plane through a pro-

posed Iceberg method on a 4 loudspeakers setup on three diﬀerent reverberation time

scenarios.

Three diﬀerent simulated rooms were measured utilizing ﬁles generated via

Iceberg method to a four-loudspeaker setup, keeping the listener in the center

position. Figure 5.16 presents the acquired ITDs with Iceberg method for RT

= 0 s (blue), RT = 0.5 s (red), RT = 1.1 s (yellow) and the ITD for RT =

0 without using virtualization (i.e., reproducing through real loudspeakers) in

Chapter 5. Main Results 147

black. There were no substantial diﬀerences across the diﬀerent reverberation

times virtualized. This was expected, though, as the direct sound drives the

ITD.

5.4.1.2 Interaural Level Diﬀerence

Figure 5.17 shows the ILDs (calculated following Equation 3.7) across octave

bands for the angles around the center horizontally spaced 30º for better vi-

sualization. The ILDs were most aﬀected than the ITDs, with a substantial

reduction in ILD relative to the actual loudspeakers observed in the 2 kHz

band. The ILD values have a similar pattern and magnitude for a signiﬁcant

part of the spectrum at most angles.

Figure 5.17: Iceberg Interaural Level Diﬀerences as a function of octave-band cen-

ter frequencies; separate lines for angles of incidence.

Figure 5.18 presents the ILD for both setups: real loudspeakers and Iceberg

method in six octave bands as a function of azimuth in spaces of 15º. The

top-right corner graph shows the 2 kHz band. It shows that apart from the

positions where there is an actual loudspeaker (i.e., 0º, 90º, 180º, and 270º),

the diﬀerences are large, greater than 10 dB at some azimuth angles.

Chapter 5. Main Results 148

Figure 5.18: Iceberg ILD as a function of azimuth angle. Listener alone in the

center.

Figure 5.19 shows the absolute diﬀerence in ILD between physical loudspeakers

and virtual loudspeakers created by the Iceberg method.

Figure 5.19: Iceberg method: Absolute ILD Diﬀerences as a function of azimuth

angle.

Tu [291] measured just-noticeable diﬀerences (JNDs) in ILDs for normal-

hearing participants using pure tones at diﬀerent presentation levels. These

JNDs can be used to estimate the perception of diﬀerences between ILDs

Chapter 5. Main Results 149

obtained with physical loudspeakers and the Iceberg auralization setup to an-

alyze if the ILD diﬀerence between setups will be perceived in a given speciﬁc

frequency band. Figure 5.20 presents the values from Figure 5.19 minus the

appropriate pure-tone ILD JND values. That means that positive values that

exceeded the JND could be perceived not as intended; that is, a perceptible

ILD deviation can cause spatial distortion [38]. The 2-kHz ILDs show up to

8º divergence across most angles. ILDs in other frequency bands (1, 2, 8, and

16 kHz) also presented values that could relate to noticeable diﬀerences (up to

4 dB), but those are mostly limited to frontal ±30° angles.

The 2 kHz mismatch can be considered a ﬂaw in the reproduction system.

The eﬀect on sound localization and subjective impression of complex sounds

involving these frequencies needs further investigation as to the scale of spatial

distortion. As the ITDs and ILDs at other frequencies were relatively well

preserved in the auralized system with only real loudspeakers, it is possible

this ﬂaw at 2 kHz may have a minimal eﬀect, especially for lower frequency

stimuli. System reliability should be veriﬁed ﬁrst for stimuli with peak energy

in the 2 kHz band or tasks requiring greater localization accuracy (e.g., with

sound sources within ±30°).

Figure 5.20: Iceberg: Absolute ILD Diﬀerences over JND as a function of azimuth

angles around the central point).

Chapter 5. Main Results 150

5.4.1.3 Azimuth Estimation

The frontal azimuth angle was estimated using the binaural model by May and

Kohlrausch [182]. Each BRIR was convolved with a pink noise with a duration

of 2.9 s as input into the model. The mean of the azimuth of each ﬁle is stated

as azimuth predicted by the model. Figure 5.21 presents the angles estimated

with the May and Kohlrausch model for ﬁles auralized with the Iceberg method

for an anechoic room and virtualized over the four-loudspeaker setup (blue

curve), angles estimated for binaural ﬁles acquired without virtualization with

real loudspeakers (red curve), and the reference (dotted black).

Figure 5.21: Iceberg method: Estimated azimuth angle (model by May and

Kohlrausch [182]), HATS centered, and RT = 0s.

The model’s results are in line with the analysis of the binaural cues supporting

the assumption of the worst localization accuracy around ±30º diﬀerence (30º

and 330º in Figure 5.21). Also, the virtualized sound tends to have more

diﬃculty separating from the frontal angle (0º).

Chapter 5. Main Results 151

5.4.2 Oﬀ-Center Positions

Moving the primary listener oﬀ-center (displaced both on the x and y axes)

is proposed to measure the impact of a person’s head (and body) not being

centered – such as when not ﬁxated – on the system’s ability to render the

appropriate binaural cues.

5.4.2.1 Interaural Time Diﬀerence

Figure 5.22 presents 72 measured ITDs around the listener (5º spacing) in four

diﬀerent placements: at center and displaced forwards (y-axis) 2.5, 5 and 10

cm.

When displaced from the center position, the Iceberg method can cope with

delivering a reasonably interaural time diﬀerence in frontal displacements up

to 5 cm or considering a simultaneous misplacement lateral and frontal up

to 2.5 cm. However, compared to the center position, the error increased

dramatically with 10 cm displacement for frontal angles (around ± 45) up to

400 µs compared to the listener in the center.

Lateral displacement positions (2.5 and 5.0 cm) were also investigated. The

ITD results for these displacements presented the same trend as seen without

lateral displacement. Similar results were found when virtualizing the scenes

with a reverberation time of 0.5 and 1.1 seconds. All combination results are

presented in Figure 5.23 to improve readability.

Chapter 5. Main Results 152

Figure 5.22: Iceberg ITD as a function of frontal displacement: Centered listener

in proposed Iceberg method in a four-loudspeaker setup.

Figure 5.23: ITD Iceberg virtualized setup: Listener displacement: listener position

2.5 cm oﬀ-center in proposed Iceberg method in a four-loudspeaker setup.

ITDs were aﬀected with frontal displacements depending on the amount of

reverberation simulated. In the simulated dry condition the squared behav-

ior is present with 5 cm oﬀ center, with mild reverberation the eﬀect only

Chapter 5. Main Results 153

appears with a displacement of 10 cm and the largest reverberation tested

demonstrated the problem to virtualize sources in all oﬀ center positions. The

deviation is centered to ±45º in all conditions. Lateral movements were even

more aﬀected, as expected, delivering ITDs based on loudspeaker position (the

squared shape) and not by circular placement of virtualized sound sources on

displacements further than 3.5 cm from the center (combining the lateral and

frontal movements.)

5.4.2.2 Interaural Level Diﬀerence

Figure 5.24 presents the diﬀerence between the ILDs measured in the center

and the ILDs measured in diﬀerent positions for a dry room simulation (RT =

0 s). The lateral displacement (x-axis) is ordered as rows (top row = center,

middle row = 2.5 cm and bottom row = 5 cm to the right). The four columns

are related to the frontal displacement (y-axis) of 0 (center), 2.5, 5, and 10 cm.

Note that these are additional ILD errors to the previously discussed errors

introduced by the simulation itself (with the listener at center).

Figure 5.24: Diﬀerence in ILD as a function of azimuth angle for a HATS Br¨uel

and Kjær TYPE 4128-C in the horizontal plane through a proposed Iceberg aural-

ization method on a four-loudspeakers setup (RT = 0.0 s).

Chapter 5. Main Results 154

The ILDs are aﬀected by listener displacement mostly in the mid frequencies

and only at certain angles. Lateral displacement of 2.5 cm produces larger

interference (up to 8 dB) for the left angles 40º and 130º. In contrast, at

other angles, ILD diﬀerences are lower than 3 dB. The 5 cm displacement

also presents diﬀerences up to 15 dB at these same angles and up to 8 dB

diﬀerences contralaterally (220º and 320º). Frontal displacement follows a

similar pattern with more diﬀerences at some of the rear angles (130º and

220º). These particular diﬀerences indicate relatively low impact on ILD cues

using the Iceberg method in the simulated anechoic room (RT = 0 s). Similar

results in terms of aﬀected angles were found analyzing ILDs for the same

listener positions for simulated rooms with RT = 0.5 s Figure 5.25 and RT

= 1.1 s Figure 5.26. These conditions are closer to everyday situations. The

increased energy of the late reﬂections results in lesser magnitude diﬀerences

in ILD, indicating slightly better performance for more realistic simulations.

Figure 5.25: Diﬀerence in ILD as a function of azimuth angle for a HATS Br¨uel

and Kjær TYPE 4128-C in the horizontal plane through a proposed Iceberg method

on a four-loudspeakers setup RT = 0.5 s.

Chapter 5. Main Results 155

Figure 5.26: Diﬀerence in ILD as a function of azimuth angle for a HATS Br¨uel

and Kjær TYPE 4128-C in the horizontal plane through a proposed Iceberg method

on a four-loudspeakers setup RT = 1.1 s.

5.4.2.3 Azimuth Estimation

Using again the May et al. model, in the same setup as in Section 5.4.1.3 to

predict the localization of a sound source, Figure 5.27 presents the predicted

source locations when moving the listener along the grid positions mentioned

(x = 0, 2.5 and 5 cm; y = 0, 2.5, 5 and 10 cm).The diﬀerent RTs are represented

by the line colors in the graphs (blue = 0.0s, red = 0.5s, yellow = 1.1s). The

results indicate the system’s spatial sound accuracy is dependent of the listener

position. On the other hand the error is not dependent on the reverberation

time. Lateral movements increase the error to the side that is getting closer to

the ear, while lessened on the contra-lateral side. Frontal movements increased

the number of angles that are not delivering the correct source angle (longer

straight horizontal line around zero). The model estimates an maximum error

up to ≈30º to a listener within 3.5 cm from the center (combining Lateral anf

frontal displacement).

Chapter 5. Main Results 156

Figure 5.27: Estimated (model by May and Kohlrausch [182]) frontal azimuth angle

at diﬀerent positions inside the loudspeaker ring as function of the target angle.

The errors when comparing the angles estimated on displaced positions to the

estimated to the center position are lessened with the increment of reverbera-

tion to the majority of the angles.

5.4.3 Centered Accompanied by a Second Listener

The Binaural cues were investigated, adding a second listener to the scene

and maintaining the ﬁrst in the center (sweet spot). The second listener was

positioned in three diﬀerent lateral (x-axis) distances on the left from the

center:

• 50 cm (simulating shoulder to shoulder).

• 75 cm.

• 100 cm.

Chapter 5. Main Results 157

5.4.3.1 Interaural Time Diﬀerence

The upper row of Figure 5.28 shows the ITDs considering the setup with the

HATS alone at the center (blue line) and also with the presence of a second

listener positioned at the right side at three diﬀerent distances from the center

and the reference. The ITDs in black were computed with no virtualization as

a reference.

Figure 5.28: ITDs and absolute ITD diﬀerences as a function of angle for multiple

conﬁgurations with (colored lines) and without a second listener (black line).

There was a small diﬀerence (≈ 15 µs) as the second listener is placed at

the closest position (50 cm) considering rear and right angles. The absolute

diﬀerence has a maximum of 201 µs, equivalent to approximated 15

◦

in the

source position (see bottom row of Figure 5.28).

5.4.3.2 Interaural Level Diﬀerence

Figure 5.29 presents the diﬀerence (∆ ILD) between the ILD computed from

the BRIRs collected with and without a second listener inside the ring. The

Chapter 5. Main Results 158

panel rows top to bottom show ∆ ILDs for simulated rooms with RTs of 0, 0.5

and 1.1 s. The columns represent the diﬀerent distances between the centred

and second listener, from 50 to 100 cm, left to right.

Figure 5.29: Interaural level diﬀerences averaged octave bands as a function of

azimuth angle for a HATS Br¨uel and Kjær TYPE 4128-C in the horizontal plane

through a proposed Iceberg method on a four-loudspeakers setup.

The results show that adding a second listener impacts the ILD for the angles

shadowed by the second listener. The eﬀect is more pronounced in the higher

octave bands (8 kHz and 16 kHz), reaching approx. 14 dB, especially at

the closest and farthest distances (50 cm and 100 cm). Although there is

less impact of having a second listener at an intermediary distance (approx.

9 dB), it still produces noticeable ILD changes in the 4 kHz band. The ∆

ILD produced by the presence of a second listener is expected as a result of

natural acoustic shadowing. The analysis of ILD around the listener is also

important in the hemiﬁeld opposite to the second listener (i.e., 180-360°). The

auralization methods that rely on the full set of loudspeakers to form the sound

pressure (e.g., Ambisonics) can present noise to the side with a free path as

a physical object, preventing the sound wave from forming accordingly at the

Chapter 5. Main Results 159

center (sweet spot).

Although the Iceberg method is partially composed by ﬁrst-order Ambisonics,

which is a method that requires all loudspeakers combined to form the ap-

propriate auralization, the part that VBAP performs presents the sound only

through the indicated quadrant, not requiring the other loudspeakers to be

active. That extends the system’s robustness with a limited number of loud-

speakers and frequency limit (not being dependent on the Ambisonics order).

5.4.3.3 Azimuth Estimation

Figure 5.30 depicts the frontal azimuth angles estimated by the May et al. for

pink noise inputs of 2.9 seconds. The pink noise is convolved to the recorded

BRIRs. The BRIRs, by its time, were recorded utilizing ﬁles generated by the

Iceberg auralization method in a four loudspeaker setup. The columns in the

top row present graphs with the average estimated angle to a centered listener

accompanied by a second lister at 50 cm (light blue curve), 75 cm (red curve),

100 cm (yellow curve) according to room simulation (denoted by reverberation

time). The shaded area corresponds to the standard deviation. The top-

left graph also presents the estimated angles to the real loudspeaker condition

without virtualization (blue dotted line). Finally, the bottom row graphs show

the diﬀerences between the estimated azimuth angle and the target angle (the

estimated error).

Chapter 5. Main Results 160

Figure 5.30: Top line = Estimated localization error with presence of a second

listener; bottom line = diﬀerence to reference. Columns depicts diﬀerent RTs, line

colors diﬀerent second listener positions.

According to the model’s results, this diﬀerence reveals that the sound created

via the Iceberg method and virtualized via a four-loudspeaker setup gives

consistent localization cues even with a side listener inside the ring in the

described positions. The median error is 9.9º, and the standard deviation

is 8.8º. On the other hand, the distribution of these diﬀerences made clear

that the setup of four loudspeakers has more diﬃculty accurately presenting

the localization cues between the frontal loudspeakers, reaching up to 27º

of mismatching to these positions (45º&315º). This result is in line with

simulations from Grimm et al. [97].

The values obtained with the low number of speakers utilized (four) are in line

with the expected in the literature [97] with a equivalent pattern [84]. Although

it can raise a ﬂag for experiments needing a more precise localization represen-

tation, the Iceberg method can improve simple setups’ realism. Accordingly,

it needs to be thoroughly investigated considering subjective listening tests,

especially in the lateral angles.

Chapter 5. Main Results 161

Figure 5.31 presents the absolute diﬀerences between the estimated angles of

arrival for the centered position alone and the centered accompanied by a

second listener at 50 cm, 75 cm, and 100 cm in the three simulated rooms

tested. These diﬀerences reﬂect the estimated inﬂuence of having the second

listener inside the ring.

Figure 5.31: Absolute diﬀerence to target in estimated localization considering the

presence of a second listener and the reverberation time.

The average error presents a slight increase with the proximity of the second

listener. However, the eﬀect is less perceivable in moderate RT. That fact

suggests that the acoustic shadow is present.

An ANOVA analysis of the variance of the estimated absolute errors between

the RT and KEMAR position groups was proposed. For the distribution with

8 degrees of freedom and a number of observations equal to 30, the tabulated

value of the F Distribution of Snedecor on (p=0.05) is equal to 2.26. Thus,

values greater than the tabulated F accept the null hypothesis that there is no

signiﬁcant diﬀerence between the means of the absolute error of the groups of

angles H

: µ

= µ

. From the analysis, the results of the F statistic (presented

in Table 5.2) H

is accepted in all groups.

Chapter 5. Main Results 162

Table 5.2: One way anova, columns are absolute diﬀerence between estimated and

reference angles for diﬀerent KEMAR positions and RTs.

Source SS df MS F Prob F

Columns 746.5 8 93.3142 1.3755 0.2062

Error 21980.9 324 67.8423

Total 22727.4 332

Therefore, there is no statistical diﬀerence between the KEMAR positions for

any of the evaluated RTs. That suggests the method’s stability in this setup,

even with a second listener considering the reverberations and positions tested.

The model has diﬃculty estimating the extreme lateral locations (90º and 270),

as even the actual loudspeakers could not reach this estimation. A comparison

between the estimated angles with a listener alone in the center position ac-

quired only with actual loudspeakers (without virtualization) and the listener

in the center accompanied by a second listener acquired from virtualized ﬁles

with the Iceberg model is presented in Figure 5.32.

Figure 5.32: Estimated error to RT=0 considering the estimation of real loud-

speakers as basis.

The data analyzed in this section suggests that by observing the indicated po-

Chapter 5. Supplementary Test Results 163

sitions where the estimated diﬀerence can be signiﬁcant, although comparable

to similar methods listed in Table 2.2 experiments with equivalent require-

ments (e.g., Chapter 4) can beneﬁt from applying the Iceberg method. The

method will fairly reproduce sounds with the presence of a second listener and

increase the sense of immersiveness while reproducing spatialized sound with

only four loudspeakers. Subjective tests are needed to investigate further the

system’s spatial rendering performance.

5.5 Supplementary Test Results

A concern about virtualization processes is how reliable they are when an extra

layer of signal processing is added to the experimental setup [97, 213]. That is,

how the sound acquisition through a hearing device microphone and its signal

processing would be aﬀected by virtualization relative to a simple loudspeaker

reproduction or the real life situation [98]. This section describes a comparison

of the binaural cues with and without hearing aids. The RIRs were collected

in the same positions as presented in Section 5.4.1 and Section 5.4.2. Further,

inspired in the study of Simon et al. [276] the robustness of the virtualization

setup outside the sweet spot was evaluated.

Oticon Opn S1 Mini RITE hearing aids with open domes were coupled to

each ear of the HATS manikin (See Figure 5.33). Modern hearing devices like

these present a series of signal processing features that can aﬀect the analysis

depending on the brand or model. In order to ensure compatibility of the

results to other devices, speciﬁc features were not enabled. The devices were

programmed in the ﬁtting software to compensate for the hearing loss of the

N3 moderate standard audiogram [35]; its beamformer sensitivity was set to

omnidirectional, and the noise reduction set to oﬀ. The hearing level of the

audiogram is presented in Table 5.3. The opened domes were chosen to set the

Chapter 5. Supplementary Test Results 164

virtualization system’s most diﬃcult signal mix condition. The signal played

through the system is not attenuated as the ear is not occluded. The ampliﬁed

signal from the hearing device is received at the eardrum (microphone) 8.1 ms

after the original signal.

Figure 5.33: HATS wearing the Oticon Open-S 1 Mini RITE.

Table 5.3: Hearing Level in dB according to the proposed Standard Audiograms for

the Flat and Moderately sloping group [35].

Nº Category Frequency

250 375 500 750 1000 1500 2000 3000 4000 6000

N3 Moderate 35 35 35 35 40 45 50 55 60 65

5.5.1 Centered Position (Aided)

The system was tested by measuring the BRIR with the listener (manikin)

in the center. The calculated binaural cues are presented at incidence angles

separated by 15º at 1.35 m from the listener.

Chapter 5. Supplementary Test Results 165

5.5.1.1 Interaural Time Diﬀerence

The ITD results (See Figure 5.34) in the aided condition were very similar to

the unaided condition 5.4.1.1.The maximum absolute diﬀerence found is 170

µs, representing a mismatch around 15º on the given angle (same as previously

measured unaided ITD diﬀerence). The calculated average diﬀerence is 67 µs,

representing a diﬀerence of around 7º in localization.

Figure 5.34: Interaural Time Diﬀerence under 1 kHz at the proposed Iceberg method

as a function of azimuth angle for a HATS Br¨uel and Kjær TYPE 4128-C in the

horizontal plane wearing a pair of hearing aids in omnidirectional mode (blue line).

The red line is the ITD results with real loudspeakers (without virtualization). The

black line represents the analytical ITD values.

Figure 5.34 depicts higher diﬀerences concentrated on speciﬁc regions: angles

around ±30º to front and back. The similarity to the unaided condition is

expected as the devices are not blocking the sound wave or increasing the

path more in one ear than the other (i.e., there is only a static group delay

added to the system). Therefore the sound reaches the HATS microphones

with hearing aids proportionally as in the previous unaided condition.

Chapter 5. Supplementary Test Results 166

5.5.1.2 Interaural Level Diﬀerence

Figure 5.35 shows the eﬀect on the ILD to the centered position. Although in

higher octave bands (8 and 16 kHz) the diﬀerence between ILD on an aided

HATS with real loudspeakers to the aided HATS utilizing the Iceberg method

is a bit larger than unaided (See Figure 5.18), the eﬀect on the 2 kHz band is

considerably smaller. That can be due to the added delay in the signal, which

can diminish the possible comb ﬁltering by the Iceberg method at this speciﬁc

frequency region, especially for the angles between two loudspeakers (where

there is a larger distance between real loudspeakers and the virtualized sound

source).

Figure 5.35: Interaural Level Diﬀerences as a function of octave-band center fre-

quencies. Angles around the central point.

5.5.1.3 Azimuth Estimation

Figure 5.36 presents the angles estimated using the May and Kohlrausch’s [182]

model for ﬁles auralized with the Iceberg method and virtualized over the four-

loudspeakers setup (blue curve), angles estimated for binaural ﬁles acquired

Chapter 5. Supplementary Test Results 167

without virtualization with real loudspeakers (red curve), and the reference

(dotted black). The presented model’s result is in line with the analysis of the

binaural cues supporting the assumption of the worst localization accuracy

around ±30º (30º and 330º in Figure 5.21). Some diﬀerences bigger than

the standard deviation are noted between diﬀerent RT, especially close to the

lateral angles (90º and 270º). The results suggest that the added reverberation

can negatively impact on localization accuracy.

Figure 5.36: Iceberg method: Estimated azimuth angle (model by May and

Kohlrausch [182]), HATS centered and aided.

Also, according to the ﬁgure, the virtualized sound tends to have more diﬃculty

separating from the frontal angle (0º) denoted by the ﬂat lines from 30º up

to 340º. Figure 5.37 depicts the boxplot diagram of the absolute diﬀerences

grouped by RT.

An ANOVA analysis of the variance of the estimated absolute errors between

the RT and position groups was proposed. For the distribution with 2 degrees

of freedom and a number of observations equal to 30, the tabulated value of

the F Distribution of Snedecor on (p=0.05) is equal to 3.32.

Chapter 5. Supplementary Test Results 168

Figure 5.37: Absolute diﬀerence to target in estimated localization in aided condi-

tion in aided condition considering diﬀerent RTs.

Thus, values greater than the tabulated F deny the null hypothesis that there

is no signiﬁcant diﬀerence between the means of the absolute error of the

groups of angles H

: µ

= µ

. From the analysis, the results of the F statistic

(presented in Table 5.4 H

is rejected and the hypothesis alternative H1 : µ

̸=

is accepted (F =5.68).

Table 5.4: One way anova, columns are absolute diﬀerence between estimated and

reference angles for diﬀerent positions and RTs.

Source SS df MS F Prob F

Columns 520.77 2 260.386 5.68 0.0045

Error 4947.29 108 45.808

Total 5468.06 110

To identify in which sets of means the discrepancy is statistically signiﬁcant,

Tukey’s multiple comparison test was performed and the result is shown in

Figure 5.38.

Chapter 5. Supplementary Test Results 169

Figure 5.38: Tukey test to compare means in aided condition. Group mean in RT

1.1s presented signiﬁcant diﬀerence from mean in group RT 0.0 s

This reﬂected a trend towards an increase in the estimated location error when

there is signal ampliﬁcation through the hearing aid, which did not occur in

the similar condition without the aid seen in Section 5.4.1.3.

5.5.2 Oﬀ-center Positions (Aided)

The listener was moved from the center position to simulate a displaced test

participant wearing hearing aids. The BRIRs were measured in the positions

described in Section 5.3.3, and the results were analyzed in this section.

5.5.2.1 Interaural Time Diﬀerence

Figure 5.39 presents the ITD for the diﬀerent angles around the listener as the

listener is displaced to diﬀerent positions according to the speciﬁed grid.

Chapter 5. Supplementary Test Results 170

Figure 5.39: Interaural Time Diﬀerences as a function of octave-band center fre-

quencies. Angles around the central point.

when it moves 5 cm to front it starts to blur more the correct ITD for the

frontal angles. Especially around ±45 degrees in the frontal hemisphere where

the ITD indicates that the sound is coming from 90º or 0º angles. Further

than this distance, also the rear ±45 are aﬀected, pointing to the break of the

panning illusion. Compared to the unaided condition (Section 5.4.2.1), this

condition is slightly more sensitive to displacements

Although the ITD analysis is angle-dependent, the results in the Table 5.5

indicates that the displacement limitations can be overall mapped to indi-

cate the maximum distance. Table 5.5 shows the maximum ITD diﬀerence

according the displacement. Although the ITD analysis is angle-dependent,

the results The maximum value diﬀerence can indicate the tendency of the

ITD shape to be squared, representing no virtualization. That may help to

identify displacement limitations can be overall mapped to indicate the maxi-

mum distance. The squared behavior occurs when the sound of one individual

speaker is the main pressure contribution, arriving too early to one of HATS

Chapter 5. Supplementary Test Results 171

ears because of the HATS’s position.

Table 5.5: Maximum ∆ITD relative to the center position according to displacement,

lines refer to lateral displacement and columns refer to frontal displacement.

RT = 0.0 s

Displacement [cm] 0.0 2.5 5.0 10.0

0.0 0 [µs] 88 [µs] 182 [µs] 374 [µs]

2.5 233 [µs] 239 [µs] 364 [µs] 472 [µs]

5 317 [µs] 353 [µs] 399[µs] 566 [µs]

RT = 0.5 s

Displacement [cm] 0.0 2.5 5.0 10.0

0.0 0 [µs] 97 [µs] 229 [µs] 386 [µs]

2.5 213 [µs] 157 [µs] 313 [µs] 472 [µs]

5 317 [µs] 282 [µs] 389 [µs] 566 [µs]

RT = 1.1 s

Displacement [cm] 0.0 2.5 5.0 10.0

0.0 0 [µs] 140 [µs] 299 [µs] 341 [µs]

2.5 236 [µs] 310 [µs] 372 [µs] 437 [µs]

5 283 [µs] 372 [µs] 380 [µs] 520 [µs]

In this case frontal displacements up to 2.5 centimeters are not presenting the

square behavior and a maximum ∆ITD, of 140 µs (RT= 1.1 s), considering

the centered position as a reference. Lateral movements are more aﬀected,

starting to present the squared behavior in the transition angles between the

rear loudspeaker and the right angle (230º) and the right loudspeaker and the

front (310º). This pattern seems not be RT dependent, which is expected due

to ITD’s nature.

5.5.2.2 Interaural Level Diﬀerence

Figure 5.40 presents the ILD, considering the simulation of an anechoic envi-

ronment (RT = 0s), on 24 angles around the listener as the listener is displaced

to diﬀerent positions according to the speciﬁed grid. Compared to the normal

condition, although it presents the same pattern, the aided condition has lesser

diﬀerences between ILDs across more angles and frequencies.

Chapter 5. Supplementary Test Results 172

Figure 5.40: Diﬀerence in ILD as a function of azimuth angle for a B & K 4128-C.

Iceberg method, horizontal plane in a 4 loudspeakers setup (RT = 0.0 s).

The diﬀerences were also lessened as the RT increased, as can be seen in

Figures 5.41 and 5.42. This result shows that increasing the reverberation can

positively aﬀect the ILD error in oﬀ center positions (reducing the diﬀerences

to the ILD in the center).

Figure 5.41: Diﬀerence in ILD as a function of azimuth angle for a HATS Br¨uel

and Kjær TYPE 4128-C in the horizontal plane through a proposed Iceberg method

on a four-loudspeaker setup RT = 0.5 s.

Chapter 5. Supplementary Test Results 173

Figure 5.42: Diﬀerence in ILD as a function of azimuth angle for a HATS Br¨uel

and Kjær TYPE 4128-C in the horizontal plane through a proposed Iceberg method

on a four-loudspeakers setup RT = 1.1 s.

Chapter 5. Supplementary Test Results 174

5.5.2.3 Azimuth Estimation

Figure 5.43 presents the estimated azimuth angle [182] according to the posi-

tion of the listener. The diﬀerent RTs are represented by the line colors in the

graphs (blue = 0.0s, red = 0.5s, yellow = 1.1s). The results demonstrate that

the Iceberg method present less accuracy in the reproducing sound in frontal

angles, especially ±30

◦

and ±330

◦

. The lateral discrepancy is smaller and also

noted with real loudspeakers, what can imply that the model used has some

diﬃculty to assess that region.

Figure 5.43: Estimated frontal azimuth angle at diﬀerent positions inside the loud-

speaker ring as function of the target angle (aided condition). Model by May and

Kohlrausch [182].

According to the model’s results, to an aided listener, the localization error

is up to 30 degrees within a frontal or lateral displacement of 5 centimeters.

In the case of 10 cm of displacement, the virtualization will fail, presenting

the squared behavior on the contralateral side. The increase of reverberation

tends to maintain the maximum error magnitude, although increasing the

spread to more angles. That means the lateral side close to the loudspeaker

will present the sound source position less in the desired position but more in

Chapter 5. Supplementary Test Results 175

the loudspeaker’s physical position. Medium reverberations are less aﬀected

by displacement, meaning extreme cases should drive extra care with listener

positioning.

Chapter 5. Discussion 176

5.6 Discussion

This chapter proposed a new hybrid auralization method (Iceberg) for a vir-

tualization setup composed of 4 loudspeakers at a 1.35 m distance from the

center. This setup is a relatively limited one intended as a feasible alternative

to the much more expensive and complicated arrangements proposed and used

in the reviewed literature (See Section 2.3).

The innovation factor of the Iceberg method is the usage of a room acoustic

parameter, called center time, used to compute the transition point between

early and late reﬂections. The Iceberg’s channel mixing and distribution au-

tomation are generalized to any RIR collected or converted in Ambisonics’ ﬁrst

order.

Implemented in MATLAB, the Iceberg auralization algorithm can generate

.wav ﬁles that were virtualized in a setup with four loudspeakers (90-degree

spacing around the listener). Three simulated sound scenarios were predeﬁned

and simulated using acoustic modeling software generating RIRs in Ambisonics

format. The setup provided appropriate reverberation times even when the

listener was away from the center position. Regarding binaural cues, in the

optimal position, the maximum deviation in ITD was 170 µs, corresponding

to a shift of approximately 15º for sources around ± 30º in front and back of

the centered listener. The considerable distance between loudspeakers is the

most likely cause of this deviation.

In contrast with Chapter 3, the Iceberg method could not reproduce the ITDs

with the same accuracy as VBAP in the sweet spot position. However, it

presented a better performance than Ambisonics. The high accuracy of VBAP

can be attributed to the number of loudspeakers, 24, which lessen their physical

distance, therefore, its maximum error. However, even with the bigger number

Chapter 5. Discussion 177

of loudspeakers, Ambisonics was truncated at ﬁrst-order, thus not having the

beneﬁt of more sound sources.

There were also deviations in ILD, mainly at the same angles. However, the

ILD deviations were most signiﬁcant in the 2 kHz octave band. The actual

eﬀects of this diﬀerence in signals that encompass these frequencies should

be further investigated in validation tests. The ILDs also denoted patterns

with better representation through VBAP than ﬁrst-order Ambisonics in Sec-

tion 3.3.2.2. The Iceberg method with four loudspeakers presents a pattern

with ILDs closer to actual loudspeakers than pure Ambisonics, but again not as

accurate as VBAP in 24 channels. This is characterized mainly to a diﬀerence

in the 2 kHz octave band. That needs further investigation and consideration

in experiments requiring ILDs accuracy at that frequency band.

Overall the results for the binaural cues reproduction via the Iceberg method

in four loudspeakers are better than a pure Ambisonics ﬁrst-order but worse

than VBAP (considering 24 loudspeakers). Therefore the Iceberg method can

be considered an option when the number of loudspeakers is limited or the need

for a sense of realism is higher. The Iceberg method combines relative accuracy

with a sense of immersion. The maximum estimated localization uncertainty

was around 30 degrees to the Iceberg method in the minimal conﬁguration

of four loudspeakers. The diﬀerent amounts of reverberation tested did not

impact the results. Although the estimated localization was imperfect, the

method’s performance was in line with similar VBAP implementations [97].

The results were similar to the aided condition, with ILDs indicating better

cue reproduction in the 2 kHz octave band. This improvement was not trans-

lated into a better-estimated angle, getting about the same results. A slight

variation was identiﬁed, and a statistically signiﬁcant diﬀerence was found be-

tween diﬀerent RTs, especially in the lateral angles. This deviation needs to

be further evaluated with other models and also with subjective validation,

Chapter 5. Discussion 178

especially as the model results presented unexpected results in these angles for

non virtualized sound sources.

A second listener was introduced at the side of the primary listener while

maintaining the listener in the optimal position to simulate a condition where

there is a need for social interaction or presence in a test. In this case, the

binaural cues provided by the Iceberg auralization method virtualized in a four

loudspeakers setup were compared to a baseline without virtualization (actual

loudspeakers). Also, the model of May and Kohlrausch [182] was applied to

predict frontal localization accuracy. Three distances were tested with the

three simulated rooms (diﬀerent RTs). There was the expected acoustical

shadow at angles blocked by the second listener but not to the remaining

sound source locations around the listener. That can be considered a measure

of rendering robustness; the second listener did not break the virtualization of

binaural cues by scrambling the sound pressure summation.

Regarding sub-optimal positions to unaided HATS, the Section 5.4.2 presented

surprising results. The the virtualized eﬀect was aﬀected diﬀerently to the dis-

placed HATS according to the amount of RT. In the dry RT (0 s) displacements

up to 5 cm when moving forward did not present the undesired eﬀect. The

mild reverberation (0.5 s) got the undesired eﬀect only with 10 cm from the

center and the large RT up to 5. The large reverberation (1.1 s) was heavily af-

fected presenting the squared behavior on all oﬀ center displacements. Lateral

movements were aﬀected in similar way for all the RTs tested, presenting the

eﬀect on displacements further than 3.5 cm from center. The ILDs presented

the shadowing eﬀect as expected, increasing the distortions with the distance

and reducing it with the increase of reverberation. The combination came to

an estimated azimuth angle in practice not dependent of RT and an error oﬀ

≈ 30º with displacements up to 2.5 cm. As the displacement from center in-

creases the maximum estimated error increases but also moved, meaning that

Chapter 5. Discussion 179

the virtualization is aﬀected but still would produce a virtualized eﬀect. In

the aided condition (Section 5.5.2) the oﬀ-center ITDs indicated a maximum

frontal displacement on the Iceberg method under 10 cm from the center in

unaided condition and under 5 cm in aided condition. The ILDs were also im-

pacted by distance, but to a lower extent, while the diﬀerent RT aﬀected less

the ITDs and more the ILDs. The ILDs present a smearing behavior, lowering

the error, with higher RT. That behavior suggests an equivalent compensation

on the error predicted by the model. Within that distance limit, the maximum

error predicted was around 30 degrees for all RTs, agreeing at the end with

the non-aided condition.

When the listener is away from the center, the Iceberg method virtualized using

the four-loudspeaker setup increases the deviations in binaural cues compared

to the cues at those sound-source angles, with a near complete loss of gradient

in cues (i.e., either zero or extreme values) occurring when the listener was

10 cm in front of the center position. ITDs for this condition revealed minor

diﬀerences across the tested reverberation conditions. The values indicated

that the ﬁles created by the method and reproduced on the four loudspeak-

ers conﬁguration produce similar ITDs as the baseline condition, having the

week point in the 30 degrees. The absolute ITD values align with similar ex-

periments found in the literature for VBAP conﬁgurations without a second

listener [241]. The acoustic shadow is indicated by an increase of the Delta

ITD diﬀerence around 270 degrees (left side), especially to the closest position,

similar to the ﬁnding with pure VBAP Chapter 3. Also, the diﬀerence in ILDs

(∆ILDs) showed that the presence is well captured in higher frequencies. All

RTs conditions and positions demonstrated to capture diﬀerences in ILD to the

left side of the mannequin. A ∆ILD is expected as a result of natural acoustic

shadowing produced by the presence of a second listener. The beneﬁt of the

Iceberg method is that the VBAP is not limited in frequency by aliasing in the

higher frequencies and does not require all to be active loudspeakers simultane-

Chapter 5. Discussion 180

ously as pure Ambisonics. The way the division is done in the Iceberg method

brings the Ambisonics’ responsibility to the time domain, deﬁning the method

as more natural to physical presence between loudspeakers and the listener.

That extends the system’s robustness with a limited number of loudspeakers

and frequency limit (not being dependent on the Ambisonics order).

The predicted error for a second listener compared to the Iceberg baseline

condition (listener centered alone). The method presents a deviation of around

10 degrees in all RTs when the second listener is in the shoulder-to-shoulder

situation, the closest position (50 cm). As the diﬀerence is at the second

listener position, it is possible to argue that the Iceberg energy balance is

advantageous, not entirely depending on the four loudspeakers’ summation.

Therefore, compared to VBAP or Ambisonics, the Iceberg method is a suitable

option in terms of localization that adds the beneﬁt of immersiveness in a

modest hardware.

5.6.1 Subjective impressions

The auralization was compared by the author and his supervisor to VBAP and

Ambisonics in subjective listening sessions. The experiment was not performed

systematically, as the Covid-19 emergency rules imposed a series of restrictions

and these impressions are the initial opinion. The speech signals were auralized

via Iceberg, VBAP, and Ambisonics and reproduced in an anechoic room. The

rooms were simulated in Odeon software v.12 with reverberation time equiva-

lent to 0.5 and 1.3 seconds. Both agreed that the sound direction from VBAP

is easily identiﬁable but with poor immersiveness, as all the reverberation came

from a speciﬁc side (2 loudspeakers). Ambisonics oﬀered a more immersive ex-

perience with all loudspeakers active simultaneously, but the localization was

Chapter 5. Discussion 181

very diﬃcult; a ”blurred” position seems to be a trending description. The Ice-

berg system provided a sound localization close to VBAP while maintaining

the immersiveness.

The Iceberg method, upon a trade-oﬀ on spatial localization, allows for the

reproduction of sounds that can be easily manipulated regarding sound-source

direction, sound pressure level, reverberation time, and simultaneous sound

sources. That makes it possible to create or reproduce speciﬁc virtual sound

scenarios with high reproducibility. Thus, researchers can conduct auditory

tests with increased ecological validity in spaces that usually do not count

with numerous loudspeakers, as is common in clinics, universities, and small

companies. Notwithstanding these beneﬁts, some limitations challenged the

method with a small number of loudspeakers. These limitations impose some

constraints on its use in terms of the spatial localization of sound sources.

5.6.2 Advantages and Limitations

A fundamental advantage of the proposed Iceberg method is the minimum

number of loudspeakers required (four). Furthermore, its compatibility with

any RIR in low- or high-order Ambisonics already collected. That is possi-

ble as an RIR in HOA can be easily scaled down to ﬁrst-order Ambisonics

and its sound spatial properties composed with any given sound via the al-

gorithm [237]. Furthermore, an essential part of the method’s deﬁnition and

an additional advantage is the automation of the deﬁnition of the amount of

energy from the RIR that corresponds to the speciﬁc auralization technique.

That automation performed by the Central Time room acoustics parameter

allows a smooth transition between the direct sound and early reﬂections por-

tion and the late reﬂections of the RIR, resulting in a potentially more natural

sound while maintaining control over the incidence direction.

Chapter 5. Concluding Remarks 182

The auralization method is designed for a virtualization setup of 4 loudspeak-

ers. However, it is possible to use it with more loudspeakers, reducing the

eventual limitations on spatial accuracy. Furthermore, although not within

the scope of this thesis, the method, using the VBAP technique, would allow

the possibility for dynamically moving sound sources around the listener.

5.6.3 Study limitations and Future Work

The initial aim of this study was to investigate the correlation between objec-

tive parameters related to spatial sound, particularly those psychoacoustically

motivated by auralization methods, and subjective responses to these meth-

ods. However, due to the Covid-19 pandemic, tests with participants were

not possible due to the risk of infection as mandated by government rules.

As a result, the study is limited to verifying objective parameters. Therefore,

section 5.5 was included to explore the system capabilities within a relevant

context for hearing research, although without subjective tests involving par-

ticipants. In future work, structured validation with participants would be

of value to the ﬁeld, allowing for adjustments and the measurement of the

eﬀectiveness of this method in real-world auditory tests. Additionally, future

implementations of this method could include improvements such as guided

sound source movements around the listener, with simultaneous updates of

VBAP and Ambisonics weights deﬁned by time constants, and the ability to

pan with intensity using techniques such as Vector-Based Intensity Panning

(VBIP), which could be tailored to speciﬁc cases with diﬀerent loudspeaker ar-

rangements or stimuli frequency content and potentially merged with VBAP

depending on the type of stimuli and speciﬁc frequencies.

Chapter 5. Concluding Remarks 183

5.7 Concluding Remarks

Tests that require hearing aids can be performed, considering some constraints,

utilizing the proposed Iceberg method. These tests aimed to verify the impact

of the auralization method through a simple setup (four loudspeakers) to the

virtualized spatial impression by analyzing the binaural cues and their devi-

ations from actual sound-sources loudspeakers. This is an important step,

although not discounting the importance of validation with test participants.

To a centered listener, the veriﬁed deviation in binaural cues presented limi-

tations of around 30º degrees in localization (through ITD) with reasonably

matching ILDs. The system’s reliability is compromised as the listener is

moved out from the sweet spot, but less so than when unaided, possibly due

to comb ﬁltering or the addition of compression into the signal path. Small

movements up to 2.5 cm generated errors within a JND, meaning they likely

would not be perceived as distortions or artifacts. Thus, tests with people that

require sound sources positioned in spaces larger than 30º can beneﬁt from this

Iceberg method that incorporates spatial awareness and immersiveness.

Chapter 6

Conclusion

Throughout the course of this study, a new auralization method called Iceberg

was conceptualized and compared to well-known methods, including VBAP

and ﬁrst-order Ambisonics, using objective parameters. The Iceberg method is

innovative in that it uses TS to ﬁnd the transition point between early and late

reﬂections in order to split the Ambisonics impulse responses and adequately

distribute them. VBAP is responsible for localization cues in this proposed

method, while Ambisonics contributes to the sense of immersion. In the center

position, the Iceberg method was found to be in line with the localization

accuracy of other methods while also adding to the sense of immersion. Also,

a second listener added to the side did not present undesired eﬀects to the

auralization. Additionally, it was found that virtualization of sound sources

with Ambisonics can implicate limitations on a participant’s behavior due to

its sweet spot in a listening-in-noise test. However, these limitations can be

circumvented and extended to Iceberg, resulting in subjective responses that

align with behavioral performance in speech intelligibility tests and increasing

the localization accuracy.

184

Chapter 6. Iceberg 185

6.1 Iceberg

In the previous chapter, we conducted a thorough analysis comparing the

performance of the Iceberg method to the results presented in Chapter 3 and

the relevant literature in Chapter 2. This comparison included evaluating

the Iceberg method’s performance at the center position, at various oﬀ-center

positions, and in the presence of a second listener. The results showed that

the Iceberg method was able to provide the designed overall reverberation

times of 0 seconds, 0.5 seconds, and 1.1 seconds across all measured positions.

Additionally, the diﬀerences between the reverberation times were below the

JND 5% threshold.

When comparing values to the ones obtained with a HATS in the center with-

out virtualization, it is noteworthy that the Iceberg method uses 20 fewer loud-

speakers than this VBAP and Ambisonics conﬁguration. The Iceberg method

exhibited lower accuracy in reproducing ITDs at the sweet spot position than

VBAP, but it performed better than ﬁrst-order Ambisonics. We also observed

detrimental deviations in ILDs, with values exceeding 4 dB, particularly at

the same angles as the ITDs. The most signiﬁcant ILD deviations occurred

in the 2 kHz octave band, which could inﬂuence the perceived localization

accuracy. Further investigation through validation tests is necessary to fully

understand the extent of these diﬀerences between the methods. Regarding

overall binaural cue reproduction, the Iceberg method using four loudspeakers

was superior to pure ﬁrst-order Ambisonics but less accurate than VBAP with

24 loudspeakers. The Iceberg presented a maximum estimated localization er-

ror of around 30 degrees for angles plus minus 40 degrees from the center while

the listener is centered. Although this magnitude matches the similar methods

in Table 2.2, the binaural cues were pointed to a lower estimate (around 15

degrees). Therefore further studies with perceptual evaluation are highly en-

couraged. In the Aided condition, we observed that the ITD was not aﬀected

Chapter 6. Iceberg 186

at the center position, and the ILD was closer to the VBAP condition with

24 loudspeakers. However, this improvement was not reﬂected in the model

estimate, which still showed maximum deviations of around ±30

◦

At oﬀ-center positions, the Iceberg method showed slight variations in localiza-

tion estimates, particularly in lateral angles, which were found to be statisti-

cally signiﬁcant when comparing diﬀerent reverberation times. This variation

is likely due to the method’s spatial limitation, known as the sweet spot, as

discussed in the Chapter 2. When the reverberation time was 0 s or 1.1 s,

the sweet spot was more limited in terms of displacement from the center (up

to 3.5 cm). This means that these conditions were more prone to breaking

virtualization when sound sources were virtualized on the contralateral side of

the displacement. In contrast, the mild condition (0.5 s) maintained this up

to 5 cm. A sweet spot is generally smaller in ﬁrst-order Ambisonics compared

to VBAP with a 24 loudspeaker setup, as identiﬁed in Chapter 3. However,

it is important to note that objective parameters may not always correspond

directly to subjective impressions. Despite this, the Iceberg method with four

loudspeakers was found to perform similarly to VBAP (with 24 loudspeakers)

in terms of binaural cue reproduction. The model estimates also showed that,

within a combined displacement of up to 3.5 cm in both lateral and frontal

directions, the maximum error would be less than 30 degrees, indicating the

presence of virtualization (i.e., the sound being physically composed of more

than just the nearest speaker). It is therefore recommended to evaluate this

deviation further using other models and subjective validation tests.

The results in Section 5.4.3.3, the condition with the listener in the center,

showed that the presence of a second listener did not negatively aﬀect the

performance of the Iceberg method in all conditions of reverberation tested.

No statistical diﬀerence in the means of estimated error was identiﬁed when

considering the three RT conditions and the three KEMAR positions. The

Chapter 6. General Discussion 187

binaural cues errors followed the same trend as the Alone version, meaning that

ITDs pointed to an error around 15 degrees, but with ILDs having absolute

values with diﬀerences exceeding 4 dB (JND), which can probably explain the

30º error estimated by the model in the worst position (i.e., the angle of the

virtualized sound source at ±45º). Based on these results, the Iceberg method

can be viable for virtualization setups with limited loudspeakers or when a

higher sense of realism is desired.

6.2 General Discussion

In this work, we explored the use of auralization methods in hearing research as

a means of improving the ecological validity of acoustic environments. The use

of virtualized sound ﬁelds has become increasingly popular in laboratory tests.

However, it is essential to understand the limitations of these methods in order

to ensure unbiased results [97]. Our literature review (Chapter 2) identiﬁed

the need for auralization methods that can be implemented in smaller-scale

setups, and our initial evaluations focused on the spatial accuracy of several

fundamental auralization methods, as well as their potential use in tasks in-

volving multiple listeners. A collaborative study allowed us to test one of these

techniques with real participants, and our ﬁndings highlighted both the limita-

tions and potential improvements of using Ambisonics for conducting listening

eﬀort tests. Based on this experience and our knowledge of room acoustics and

auralization, we proposed a new hybrid method called Iceberg, which combines

the strengths of Ambisonics and VBAP and can be implemented using just four

loudspeakers. This proposed method oﬀers a low-cost option for auralization

that could increase its adoption among researchers worldwide.

In Chapter 3, the VBAP and Ambisonics auralization methods were objectively

characterized and compared in terms of binaural cues for the center and oﬀ-

Chapter 6. General Discussion 188

center positions. This investigation provided a foundation for combining the

methods and further highlighted the strengths of each technique: localization

in VBAP and immersiveness in Ambisonics. Objective parameters extracted

from BRIRs and RIRs were examined for a single listener and in the presence

of a second listener in the room. The results showed that the presence of

a second listener did not signiﬁcantly impact the performance of VBAP. At

the same time, Ambisonics was less eﬀective in reproducing the examined

cues, especially with a second listener present. This information was crucial

in developing the proposed Iceberg auralization method, which combines the

strengths of both VBAP and Ambisonics to create a hybrid method suitable

for use with simple setups such as four loudspeakers.

The results of the collaborative study described in Chapter 4 demonstrate the

feasibility of using a virtualization method to deliver a hearing test with a

certain level of spatial resolution and immersion across diﬀerent room simula-

tions and signal-to-noise ratios. This study suggests that virtualization meth-

ods have the potential to provide realistic acoustic environments for hearing

tests, allowing researchers to vary the acoustic demands of a task and po-

tentially improve ecological validity. Additionally, the signiﬁcant correlation

between participants’ subjective perception of eﬀort and their speech recogni-

tion performance highlights the importance of considering listening eﬀort in

hearing research. However, the limitations and potential solutions identiﬁed in

this study also highlight the need for further investigation into virtualization

methods in hearing research, including developing new auralization methods

that address these limitations.

In Chapter 5, we presented the development of a new auralization method

called Iceberg, which was designed to be compatible with small-scale virtu-

alization setups using only four loudspeakers. Previous hybrid methods that

combine Ambisonics and VBAP have been developed, but the innovative as-

Chapter 6. General Discussion 189

pect of the Iceberg method is its approach to handling and combining the dif-

ferent methods to virtualize sounds while delivering appropriate spatial cues.

This feature is achieved by identifying a transition point in the RIR using the

Central Time parameter from the omnidirectional channel of an Ambisonics

RIR. This automated process allows the user to input any Ambisonics RIR,

along with the desired presentation angle(s) and sound ﬁle(s), to be auralized

using the VBAP and Ambisonics methods merged into a ﬁnal multi-channel

.wav ﬁle for presentation over a four-loudspeaker system. One of the beneﬁts

of this approach is that it does not require any additional parameters, such as

those generated by a simulation program, and can be used with any Ambison-

ics RIR, including those in higher-order format that must be converted to an

appropriate order for the number of loudspeakers. Overall, the development of

the Iceberg method illustrates the potential for adapting existing technology

to meet the needs of smaller-scale virtualization setups while still delivering

realistic spatial cues. This approach could support the broader adoption of au-

ralization in hearing research and encourage researchers to utilize virtualized

sound ﬁelds in their protocols.

6.2.1 Iceberg capabilities

The auralization method proposed in this work combines the use of Ambisonics

RIRs and VBAP to balance the acoustic energy in two spatial domains: the

perception of sound localization and the perception of immersion. This results

in a ﬁle that captures the characteristics of a given sound as if it were played

in the desired environment. The method can be reproduced with at least

four loudspeakers but is scalable to a more extensive array of loudspeakers of

any size greater than four, theoretically increasing its eﬃciency. In addition,

multiple sound sources can be virtualized and merged at presentation to create

more complex environments. The input to Iceberg includes Ambisonics RIRs

Chapter 6. General Discussion 190

corresponding to speciﬁc source-and-receptor positions and the sounds to be

virtualized, preferably recorded in (near) anechoic conditions. The method can

pan the source around the listener, as the VBAP component is independent of

Ambisonics. However, it is recommended that RIRs be generated for speciﬁc

angles when using room acoustic software to generate the Ambisonics RIRs.

One beneﬁt of this method is that it can reproduce sounds above the cut-oﬀ

frequency associated with lower-order Ambisonics due to its use of VBAP,

which is initially not frequency limited [241]. VBAP is responsible for the

delivery of both direct sound and early reﬂections. Additionally, the default

properties are deﬁned to work with normalized RIRs, enabling the researcher

to specify the sound pressure level of the auralized ﬁles.

6.2.2 Iceberg & Second Joint Listener

Testing with a second listener inside the loudspeaker ring helps illuminate the

potential for this virtualization system in diﬀerent tasks and human-interaction

situations [143, 202, 230, 234]. A system that allows these tasks and situations

needs to deliver the appropriate sound properties for the sound to be per-

ceived as coming from the intended position [97]. Ambisonics was shown to

be not eﬀective in this test, as the shadow caused by a second listener pre-

vented higher frequency spatial information from being correctly presented,

distorting the sound ﬁeld (especially in low-order Ambisonics). Vector-based

solutions can have less impact as the sound is physically formed from two (or

three in 3D setups) loudspeakers in the same quadrant. That means that the

interference will happen only at angles where the acoustic shadow of a physical

object would naturally interfere in a non-virtualized reproduction. In Chap-

ter 5 BRIRs were acquired with ﬁles generated by the Iceberg method and

reproduced via a modest setup composed of four loudspeakers in the presence

of a second listener. It could be observed that it did not disturb the sound

Chapter 6. General Discussion 191

ﬁeld, as the (primary) listener in the center position received the appropri-

ate binaural cues. The system designed to reproduce ﬁles virtualized with

Iceberg method managed to perform competitively with systems with more

loudspeakers rendering pure methods (See Table 2.2).

6.2.3 Iceberg: Listener Wearing Hearing Aids

Adding the possibility of allowing participants to use hearing devices is an-

other crucial step in making auditory tests with auralized ﬁles accessible to

more researchers [134, 144]. It has been observed that hearing aid signals can

inﬂuence the intelligibility and clarity of speech in virtualized sound ﬁelds [7,

97, 99, 103, 161, 188, 213, 276]. When the hearing aid signals are not appro-

priately aligned with the characteristics of the virtualized sound ﬁeld, listeners

may struggle to comprehend spoken words or sentences [98]. This issue can be

exacerbated when the virtualized sound ﬁeld includes noise or other distrac-

tions that can interfere with speech perception or when the hearing aid signals

fail to amplify or enhance the speech signal to an adequate degree [98, 137].

Suppose the hearing aid signals are not correctly capturing the sound ﬁeld and,

therefore, not correcting it to the individual needs and preferences of the lis-

tener. In that case, the listener may experience diﬃculty using the virtualized

sound ﬁeld comfortably and eﬀectively. Swept signals were auralized by the

Iceberg method, played through the system, recorded with a manikin wearing

hearing aids, and deconvolved. The resulting BRIRs were analyzed in terms of

binaural cues and compared to the same signals from actual loudspeakers. The

localization error was estimated by May and Kohlrausch’s probabilistic model

for robust sound source localization based on a binaural auditory front end.

This model estimates the location of a sound source using binaural cues such as

interaural level diﬀerences and interaural time diﬀerences extracted from the

signals received by the two ears. By combining these cues in a probabilistic

Chapter 6. General Conclusion 192

framework, the model can robustly estimate the location of the sound source,

even in noisy or distracting environments. Evaluation of the model suggests

its potential for use in practical contexts such as in hearing aids or virtual re-

ality systems. Results obtained using the Iceberg method with an aided HATS

showed similar performance to the unaided results with the listener positioned

in the sweet spot, indicating suitable performance (see Section 5.5).

6.2.4 Iceberg Limitations

The virtualization system playing ﬁles auralized with the Iceberg method has

been found to be less eﬀective outside of the sweet spot, as the binaural cues

are not correctly rendered. This mismatch, which occurs for more than 2.5 cm

displacements, can be mitigated by keeping the listener centered in the virtu-

alized sound ﬁeld. While this is a signiﬁcant limitation, the method can still

be applied with simple measures such as a modest head restraint, reducing the

setup requirements compared to other classical methods. One major limitation

of the Iceberg method is its spatial resolution capabilities. It is recommended

for scenarios with a minimum of 30º of separation between sound sources (it

can be lower if closer to loudspeakers, although it should be checked for the

error distribution). Furthermore, the distance to the sound source should be

equal to the radius of the loudspeaker array, as Ambisonics and VBAP can-

not deﬁne sources inside the array. VBAP can only pan between physical

sound sources. These limitations should be considered when using the Iceberg

method to create virtualized sound ﬁelds.

Chapter 6. General Conclusion 193

6.3 General Conclusion

As computational capacity increases, using more complex and natural sound

scenarios in auditory research becomes feasible and desirable. This technology

allows for testing new features, sensors, and algorithms in controlled condi-

tions with increasing realism and ecological validity. Even clinical tests can

beneﬁt from auralization, allowing for investigations in diﬀerent scenarios with

varying acoustics (e.g., in a speech-in-noise test). The spatial-cue performance

of the Iceberg auralization method, reproducing ﬁles through a system of four

loudspeakers, is mainly suﬃcient for these types of tests. It is essential to un-

derstand the constraints of auralization methods, Iceberg included, which are

tied to the virtualization setup and should be chosen by researchers based on

their needs and the available hardware. However, utilizing the Iceberg, virtu-

alization can be conducted by auditory research groups that cannot aﬀord or

house expensive anechoic chambers with tens or hundreds of loudspeakers and

sophisticated hardware and need more freedom than using headphones. The

method presented in this work serves as an additional tool for researchers to

consider.

6.4 Main Contributions

In this work, we have presented a novel auralization method called Iceberg,

designed to create virtualized sound scenarios for use in auditory research. The

main contributions of this work are:

1. The development of a hybrid auralization method that combines two

psychoacoustic virtualization methods to balance the energy of an RIR

and output a multi-channel ﬁle for presentation.

Chapter 6. Main Contributions 194

2. The implementation of an eﬀective, simple, and partially automated au-

ralization method that allows for the creation of reasonably realistic vir-

tualized sound scenarios with a modest setup.

3. The exploration of the use and limitations of auralization methods in

auditory research, including the suggestion that the Iceberg method has

the potential to be a helpful tool for testing new features, sensors, and

algorithms in controlled conditions with increasing realism and ecological

validity.

4. We researched the limitations and feasibility of using Ambisonics in the

context of speech intelligibility with normal-hearing listeners.

5. Identifying the potential for the Iceberg method to be applied in a range

of practical contexts, including in hearing aids and virtual reality sys-

tems.

Bibliography

[1] Aguirre, S. L. (2017). Implementa¸c˜ao e avalia¸c˜ao de um sistema de vir-

tualiza¸c˜ao de fontes sonoras (in portuguese). Master, Programa de P´os-

Gradua¸c˜ao em Engenharia Mecˆanica, Universidade Federal de Santa Cata-

rina.

(Cited on pages 41 and 49)

[2] Aguirre, S. L., Bramsløw, L., Lunner, T., and Whitmer, W. M. (2019). Spa-

tial cue distortions within a virtualized sound ﬁeld caused by an additional

listener. In Proceedings of the 23rd International Congress on Acoustics : in-

tegrating 4th EAA Euroregio 2019, pages 6537–6544, Berlin, Germany. ICA

International Congress on Acoustics, Deutsche Gesellschaft f¨ur Akustik.

(Cited on page 94)

[3] Aguirre, S. L., Seiﬁ-Ala, T., Bramsløw, L., Graversen, C., Hadley, L. V.,

Naylor, G., and Whitmer, W. M. (2021). Combination study 3. http:

//hear-eco.eu/combination-study-3/. (accessed: 24.11.2021).

(Cited on page 97)

[4] Agus, T. R., Akeroyd, M. A., Gatehouse, S., and Warden, D. (2009). In-

formational masking in young and elderly listeners for speech masked by

simultaneous speech and noise. The Journal of the Acoustical Society of

America, 126(4):1926–1940.

(Cited on page 40)

[5] Ahnert Feistel Media Group (2011). Ease enhanced

acoustic simulator for engineers. https://www.afmg.eu/en/

ease-enhanced-acoustic-simulator-engineers. Last checked on: Nov 28,

2021.

(Cited on page 27)

[6] Ahrens, A., Marschall, M., and Dau, T. (2017). Measuring speech intelligi-

bility with speech and noise interferers in a loudspeaker-based virtual sound

environment. The Journal of the Acoustical Society of America, 141(5):3510–

3510.

(Cited on pages 16, 42 and 52)

[7] Ahrens, A., Marschall, M., and Dau, T. (2019). Measuring and modeling

speech intelligibility in real and loudspeaker-based virtual sound environ-

ments. Hearing Research, 377:307–317.

(Cited on pages 42, 53, 99 and 191)

195

BIBLIOGRAPHY 196

[8] Ahrens, A., Marschall, M., and Dau, T. (2020). The eﬀect of spatial energy

spread on sound image size and speech intelligibility. The Journal of the

Acoustical Society of America, 147(3):1368–1378.

(Cited on page 42)

[9] Akeroyd, M. A. (2006). The psychoacoustics of binaural hearing. Interna-

tional Journal of Audiology, 45(sup1):25–33.

(Cited on pages 9 and 17)

[10] Alfandari Menase, D. (2022). Motivation and fatigue eﬀects in pupillo-

metric measures of listening eﬀort. PhD thesis, University of Nottingham.

(Cited on page 33)

[11] Algazi, V. R., Duda, R. O., and Thompson, D. M. (2004). Motion-tracked

binaural sound. Journal of the Audio Engineering Society, 52(11):1142–

1156.

(Cited on page 23)

[12] Alhanbali, S., Dawes, P., Millman, R. E., and Munro, K. J. (2019). Mea-

sures of Listening Eﬀort Are Multidimensional. Ear and hearing.

(Cited on pages 51 and 118)

[13] Alpert, M. I., Alpert, J. I., and Maltz, E. N. (2005). Purchase occasion

inﬂuence on the role of music in advertising. Journal of business research,

58(3):369–376.

(Cited on page 8)

[14] Arau-Puchades, H. (1988). An improved reverberation formula. Acta

Acustica united with Acustica, 65(4):163–180.

(Cited on page 34)

[15] Archontis Politis (2020). Higher Order Ambisonics (HOA) library.

(Cited on page 107)

[16] Arlinger, S. (2003). Negative consequences of uncorrected hearing loss—a

review. International Journal of Audiology, 42(sup2):17–20.

(Cited on pages 1 and 55)

[17] Asp¨ock, L., Pausch, F., Stienen, J., Berzborn, M., Kohnen, M., Fels, J.,

and Vorl¨ander, M. (2018). Application of virtual acoustic environments in

the scope of auditory research. In XXVIII Encontro da Sociedade Brasileira

de Ac´ustica, SOBRAC, Porto Alegre, Brazil. SOBRAC.

(Cited on pages 16 and 42)

[18] Attenborough, K. (2007). Sound Propagation in the Atmosphere, pages

113–147. Springer New York, New York, NY.

(Cited on page 10)

[19] Baldan, S., Lachambre, H., Delle Monache, S., and Boussard, P. (2015).

Physically informed car engine sound synthesis for virtual and augmented

environments. In 2015 IEEE 2nd VR Workshop on Sonic Interactions for

Virtual Environments (SIVE), pages 1–6. IEEE.

(Cited on page 20)

BIBLIOGRAPHY 197

[20] Barron, M. (1971). The subjective eﬀects of ﬁrst reﬂections in concert

halls—the need for lateral reﬂections. Journal of Sound and Vibration,

15(4):475–494.

(Cited on page 37)

[21] Barron, M. and Marshall, A. (1981). Spatial impression due to early

lateral reﬂections in concert halls: The derivation of a physical measure.

Journal of Sound and Vibration, 77(2):211–232.

(Cited on pages 9, 37 and 38)

[22] Bates, E., Kearney, G., Furlong, D., and Boland, F. (2007). Localization

accuracy of advanced spatialisation techniques in small concert halls. The

Journal of the Acoustical Society of America, 121.

(Cited on pages 47, 49 and 53)

[23] Benesty, J., Sondhi, M., and Huang, Y. (2008). Springer Handbook of

Speech Processing. Springer Handbook of Speech Processing. Springer-

Verlag Berlin Heidelberg. bibtex: Benesty2008.

(Cited on page 24)

[24] Berkhout, A. J. (1988). a holographic approach to acoustic control. Jour-

nal of the Audio Engineering Society, 36(12):977–995.

(Cited on page 31)

[25] Berkhout, A. J., de Vries, D., and Vogel, P. (1993). Acoustic control

by wave ﬁeld synthesis. The Journal of the Acoustical Society of America,

93(5):2764–2778.

(Cited on page 31)

[26] Bertet, S., Daniel, J., Parizet, E., and Warusfel, O. (2009). Inﬂuence

of microphone and loudspeaker setup on perceived higher order ambisonics

reproduced sound ﬁeld. Proceedings of Ambisonics Symposium. cited By 3.

(Cited on page 31)

[27] Bertet, S., Daniel, J., Parizet, E., and Warusfel, O. (2013). Investigation

on localisation accuracy for ﬁrst and higher order ambisonics reproduced

sound sources. Acta Acustica united with Acustica, 99:642 – 657.

(Cited on pages 31 and 77)

[28] Bertoli, S. and Bodmer, D. (2014). Novel sounds as a psychophysiological

measure of listening eﬀort in older listeners with and without hearing loss.

Clinical Neurophysiology.

(Cited on page 50)

[29] Berzborn, M., Bomhardt, R., Klein, J., Richter, J.-G., and Vorl¨ander, M.

(2017). The ITA-Toolbox: An Open Source MATLAB Toolbox for Acoustic

Measurements and Signal Processing. In 43th Annual German Congress on

Acoustics, Kiel (Germany), 6 Mar 2017 - 9 Mar 2017, volume 43, pages

222–225.

(Cited on pages 60, 107 and 138)

[30] Best, V., Kalluri, S., McLachlan, S., Valentine, S., Edwards, B., and

Carlile, S. (2010). A comparison of cic and bte hearing aids for three-

dimensional localization of speech. International Journal of Audiology,

49(10):723–732.

(Cited on page 42)

BIBLIOGRAPHY 198

[31] Best, V., Keidser, G., Buchholz, J. M., and Freeston, K. (2015). An exam-

ination of speech reception thresholds measured in a simulated reverberant

cafeteria environment. International Journal of Audiology.

(Cited on pages 42 and 52)

[32] Best, V., Marrone, N., Mason, C. R., and Kidd, G. (2012). The inﬂuence

of non-spatial factors on measures of spatial release from masking. The

Journal of the Acoustical Society of America, 131(4):3103–3110. bibtex:

best2012.

(Cited on page 33)

[33] Bidelman, G. M., Davis, M. K., and Pridgen, M. H. (2018). Brainstem-

cortical functional connectivity for speech is diﬀerentially challenged by noise

and reverberation. Hearing Research.

(Cited on page 50)

[34] Bigg, G. R. (2015). The science of icebergs, page 21–124. Cambridge

University Press.

(Cited on page 128)

[35] Bisgaard, N., Vlaming, M. S. M. G., and Dahlquist, M. (2010). Stan-

dard audiograms for the iec 60118-15 measurement procedure. Trends in

Ampliﬁcation, 14(2):113–120.

(Cited on pages 163 and 164)

[36] Blackstock, D. (2000). Fundamentals of Physical Acoustics. A Wiley-

Interscience publication. Wiley.

(Cited on page 228)

[37] Blauert, J. (1969). Sound localization in the median plane. Acta Acustica

united with Acustica, 22(4):205–213.

(Cited on page 10)

[38] Blauert, J. (1997). Spatial hearing: the psychophysics of human sound

localization. MIT press.

(Cited on pages 9, 13, 14, 16, 18, 40, 73, 93, 126 and 149)

[39] Blauert, J. (2005). Communication acoustics. Springer-Verlag Berlin Hei-

delberg, 1 edition.

(Cited on pages 2, 3, 10, 13, 20, 76 and 98)

[40] Blauert, J. (2013). The technology of binaural listening. Springer.

(Cited on pages 9, 20, 22, 33 and 76)

[41] Blauert, J., Lehnert, H., Sahrhage, J., and Strauss, H. (2000). An interac-

tive virtual-environment generator for psychoacoustic research. i: Architec-

ture and implementation. Acta Acustica united with Acustica, 86:94–102.

(Cited on pages 1 and 57)

[42] Bock, T. M. and Keele, Jr., D. B. D. (1986). The eﬀects of interau-

ral crosstalk on stereo reproduction and minimizing interaural crosstalk in

nearﬁeld monitoring by the use of a physical barrier: part 1. Journal of the

Audio Engineering Society.

(Cited on page 43)

BIBLIOGRAPHY 199

[43] Bradley, J. S. (1986). Speech intelligibility studies in classrooms. The

Journal of the Acoustical Society of America, 80(3):846–854.

(Cited on page 127)

[44] Bradley, J. S. and Soulodre, G. A. (1995). Objective measures of listener

envelopment. The Journal of the Acoustical Society of America, 98(5):2590–

2597.

(Cited on pages 36, 38 and 58)

[45] Brand˜ao, E. (2018). Ac´ustica de salas: Projeto e modelagem. Editora

Blucher, S˜ao Paulo.

(Cited on pages 16, 19, 33, 36, 38 and 128)

[46] Brandao, E., Morgado, G., and Fonseca, W. (2020). A ray tracing engine

integrated with blender and with uncertainty estimation: Description and

initial results. Building Acoustics, 28:1–20.

(Cited on page 27)

[47] Breebaart, J., van de Par, S., Kohlrausch, A., and Schuijers, E. (2004).

High-quality parametric spatial audio coding at low bitrates. Journal of the

Audio Engineering Society.

(Cited on page 57)

[48] Breebaart, J., Van de Par, S., Kohlrausch, A., and Schuijers, E. (2005).

Parametric coding of stereo audio. EURASIP Journal on Advances in Signal

Processing, pages 1–18.

(Cited on page 57)

[49] Brinkmann, F., Asp¨ock, L., Ackermann, D., Lepa, S., Vorl¨ander, M.,

and Weinzierl, S. (2019). A round robin on room acoustical simulation and

auralization. The Journal of the Acoustical Society of America, 145(4):2746–

2760.

(Cited on page 21)

[50] Brinkmann, F., Asp¨ock, L., Ackermann, D., Opdam, R., Vorl¨ander, M.,

and Weinzierl, S. (2021). A benchmark for room acoustical simulation.

concept and database. Applied Acoustics, 176:107867.

(Cited on page 21)

[51] Brinkmann, F., Lindau, A., and Weinzierl, S. (2017). On the authenticity

of individual dynamic binaural synthesis. The Journal of the Acoustical

Society of America, 142(4):1784–1795.

(Cited on page 14)

[52] Brown, C. and Duda, R. (1998). A structural model for binaural sound

synthesis. IEEE Transactions on Speech and Audio Processing, 6(5):476–

488.

(Cited on page 12)

[53] Brown, V. A. and Strand, J. F. (2019). Noise increases listening eﬀort

in normal-hearing young adults, regardless of working memory capacity.

Language, Cognition and Neuroscience.

(Cited on pages 51 and 118)

BIBLIOGRAPHY 200

[54] Brungart, D. S., Cohen, J., Cord, M., Zion, D., and Kalluri, S. (2014).

Assessment of auditory spatial awareness in complex listening environments.

The Journal of the Acoustical Society of America, 136(4):1808–1820.

(Cited on page 18)

[55] Buchholz, J. M. and Best, V. (2020). Speech detection and localization

in a reverberant multitalker environment by normal-hearing and hearing-

impaired listeners. The Journal of the Acoustical Society of America,

147(3):1469–1477.

(Cited on page 42)

[56] Byrnes, H. (1984). The role of listening comprehension: A theoretical

base. Foreign language annals, 17(4):317.

(Cited on page 8)

[57] Campanini, S. and Farina, A. (2008). A new audacity feature: room

objective acustical parameters calculation module.

(Cited on page 36)

[58] Choi, I., Shinn-Cunningham, B. G., Chon, S. B., and Sung, K.-m. (2008).

Objective measurement of perceived auditory quality in multichannel audio

compression coding systems. Journal of the Audio Engineering Society,

56(1/2):3–17.

(Cited on page 73)

[59] Claus Lynge Christensen, Gry Bælum Nielsen, J. H. R. (2008). Danish

acoustical society round robin on room acoustic computer modelling. https:

//odeon.dk/learn/articles/auralisation/. Last checked on: Nov 28, 2021.

(Cited on pages 18, 27, 68, 107 and 129)

[60] Cooper, D. H. and Bauck, J. L. (1989). prospects for transaural recording.

Journal of the Audio Engineering Society, 37(1/2):3–19.

(Cited on page 22)

[61] Cubick, J. and Dau, T. (2016). Validation of a virtual sound environment

system for testing hearing aids. Acta Acustica united with Acustica.

(Cited on pages 42, 53, 56 and 120)

[62] Cuevas-Rodr´ıguez, M., Picinali, L., Gonz´alez-Toledo, D., Garre, C., de la

Rubia-Cuestas, E., Molina-Tanco, L., and Reyes-Lecuona, A. (2019). 3d

tune-in toolkit: An open-source library for real-time binaural spatialisation.

PloS one, 14(3):e0211899.

(Cited on pages 16, 20 and 121)

[63] Cunningham, L. L. and Tucci, D. L. (2017). Hearing loss in adults. New

England Journal of Medicine, 377(25):2465–2473.

(Cited on pages 1 and 55)

[64] Daniel, J. (2000). Repr´esentation de champs acoustiques, application `a

la transmission et `a la reproduction de sc`enes sonores complexes dans un

contexte multim´edia (In French). PhD thesis, University of Paris VI.

(Cited on pages 31 and 99)

[65] Daniel, J. and Moreau, S. (2004). Further study of sound ﬁeld coding with

higher order ambisonics. In Audio Engineering Society Convention 116.

(Cited on pages 30, 32, 99, 121 and 142)

BIBLIOGRAPHY 201

[66] Davies, W. J., Bruce, N. S., and Murphy, J. E. (2014). Soundscape repro-

duction and synthesis. Acta Acustica united with Acustica, 100(2):285–292.

(Cited on page 42)

[67] Dietrich, P., Masiero, B., M¨uller-Trapet, M., Pollow, M., and Scharrer,

R. (2010). Matlab toolbox for the comprehension of acoustic measurement

and signal processing. In Fortschritte der Akustik – DAGA.

(Cited on page 107)

[68] Dreier, C. and Vorl¨ander, M. (2020). Psychoacoustic optimisation of air-

craft noise-challenges and limits. In Inter-Noise and Noise-Con Congress

and Conference Proceedings, volume 261, pages 2379–2386. Institute of Noise

Control Engineering.

(Cited on page 20)

[69] Dreier, C. and Vorl¨ander, M. (2021). Aircraft noise—auralization-based

assessment of weather-dependent eﬀects on loudness and sharpness. The

Journal of the Acoustical Society of America, 149(5):3565–3575.

(Cited on page 20)

[70] Duda, R., Avendano, C., and Algazi, V. (1999). An adaptable ellip-

soidal head model for the interaural time diﬀerence. In 1999 IEEE Interna-

tional Conference on Acoustics, Speech, and Signal Processing. Proceedings.

ICASSP99 (Cat. No.99CH36258), volume 2, pages 965–968 vol.2.

(Cited on page 12)

[71] Dunne, R., Desai, D., and Heyns, P. S. (2021). Development of an acoustic

material property database and universal airﬂow resistivity model. Applied

Acoustics, 173:107730.

(Cited on page 21)

[72] Eddins, D. A. and Hall, J. W. (2010). Binaural processing and auditory

asymmetries. In Gordon-Salant, S., Frisina, R. D., Popper, A. N., and Fay,

R. R., editors, The Aging Auditory System, pages 135–165. Springer New

York, New York, NY.

(Cited on page 11)

[73] Epain, N., Guillon, P., Kan, A., Kosobrodov, R., Sun, D., Jin, C., and

Van Schaik, A. (2010). Objective evaluation of a three-dimensional sound

ﬁeld reproduction system. In Burgess, M., Davey, J., Don, C., and McMinn,

T., editors, Proceedings of 20th International Congress on Acoustics, ICA

2010, volume 2, pages 949–955. International Congress on Acoustics (ICA).

(Cited on page 31)

[74] Eyring, C. F. (1930). Reverberation time in “dead” rooms. The Journal

of the Acoustical Society of America, 1(2A):168–168.

(Cited on page 34)

[75] Farina, A. (2000). Simultaneous measurement of impulse response and

distortion with a swept-sine technique. Journal of The Audio Engineering

Society.

(Cited on page 63)

BIBLIOGRAPHY 202

[76] Farina, A., Glasgal, R., Armelloni, E., and Torger, A. (2001). ambiophonic

principles for the recording and reproduction of surround sound for music.

journal of the audio engineering society.

(Cited on pages 43 and 45)

[77] Favrot, S. and Buchholz, J. (2009). Validation of a loudspeaker-based

room auralization system using speech intelligibility measures. In Audio

Engineering Society Convention Papers, volume Preprint 7763, page 7763.

Praesens Verlag. 126th Audio Engineering Society Convention, AES126 ;

Conference date: 07-05-2009 Through 10-05-2009.

(Cited on pages 21 and 99)

[78] Favrot, S., Marschall, M., K¨asbach, J., Buchholz, J., and Weller, T.

(2011). Mixed-order ambisonics recording and playback for improving hor-

izontal directionality. In Proceeding of the audio engineering society 131st

convention. 131st AES Convention ; Conference date: 20-10-2011 Through

23-10-2011.

(Cited on page 99)

[79] Favrot, S. E., Buchholz, J., and Dau, T. (2010). A loudspeaker-based room

auralization system for auditory research. phdthesis, Technical University

of Denmark.

(Cited on pages 1, 42, 43, 45, 56, 57, 120 and 127)

[80] Fichna, S., Biberger, T., Seeber, B. U., and Ewert, S. D. (2021). Eﬀect

of acoustic scene complexity and visual scene representation on auditory

perception in virtual audio-visual environments. 2021 Immersive and 3D

Audio: from Architecture to Automotive (I3DA).

(Cited on page 42)

[81] Fintor, E., Asp¨ock, L., Fels, J., and Schlittmeier, S. (2021). The role of

spatial separation of two talkers auditory stimuli in the listener’s memory

of running speech: listening eﬀort in a non-noisy conversational setting.

International Journal of Audiology.

(Cited on page 57)

[82] Fitzroy, D. (1959). Reverberation formula which seems to be more accu-

rate with nonuniform distribution of absorption. The Journal of the Acous-

tical Society of America, 31(7):893–897.

(Cited on page 34)

[83] Francis, A. L. and Love, J. (2020). Listening eﬀort: Are we measuring

cognition or aﬀect, or both? WIREs Cognitive Science, 11(1):e1514.

(Cited on page 98)

[84] Frank, M. (2014). Localization using diﬀerent amplitude-panning methods

in the frontal hhorizontal plane. In Proceedings of the EAA Joint Symposium

on Auralization and Ambisonics 2014.

(Cited on pages 45, 49 and 160)

[85] Frank, M. and Zotter, F. (2008). Localization experiments using diﬀer-

ent 2d ambisonics decoders. In 25th Tonmeistertagung-VDT International

Convention, Leipzig.

(Cited on pages 45 and 49)

BIBLIOGRAPHY 203

[86] Franklin, W. S. (1903). Derivation of equation of decaying sound in a

room and deﬁnition of open window equivalent of absorbing power. Phys.

Rev. (Series I), 16:372–374.

(Cited on page 34)

[87] Fraser, S., Gagn´e, J. P., Alepins, M., and Dubois, P. (2010). Evaluat-

ing the eﬀort expended to understand speech in noise using a dual-task

paradigm: The eﬀects of providing visual speech cues. Journal of Speech,

Language, and Hearing Research.

(Cited on page 50)

[88] Furuya, H., Fujimoto, K., Young Ji, C., and Higa, N. (2001). Arrival direc-

tion of late sound and listener envelopment. Applied Acoustics, 62(2):125–

136.

(Cited on page 37)

[89] Gandemer, L., Parseihian, G., Bourdin, C., and Kronland-Martinet, R.

(2018). Perception of Surrounding Sound Source Trajectories in the Horizon-

tal Plane: A Comparison of VBAP and Basic-Decoded HOA. Acta Acustica

united with Acustica, pages 338–350.

(Cited on pages 40, 57 and 122)

[90] Gelfand, S. and Gelfand, S. (2004). Hearing: An Introduction to Psycho-

logical and Physiological Acoustics, Fourth Edition. Taylor & Francis.

(Cited on page 76)

[91] Gerzon, M. A. (1985). Ambisonics in multichannel broadcasting and

video. AES: Journal of the Audio Engineering Society.

(Cited on pages 23 and 58)

[92] Giguere, C. and Woodland, P. C. (1994). A computational model of the

auditory periphery for speech and hearing research. i. ascending path. The

Journal of the Acoustical Society of America, 95(1):331–342.

(Cited on page 8)

[93] Gil Carvajal, J., Cubick, J., Santurette, S., and Dau, T. (2016). Spatial

hearing with incongruent visual or auditory room cues. Scientiﬁc Reports,

(Cited on pages 16 and 42)

[94] Glasgal, R. (2001). the ambiophone derivation of a recording methodology

optimized for ambiophonic reproduction. journal of the audio engineering

society.

(Cited on page 43)

[95] Glasgal, R. and Yates, K. (1995). Ambiophonics: Beyond Surround Sound

to Virtual Sonic Reality. Ambiophonics Institute.

(Cited on page 43)

[96] Gomes, L., Fonseca, W. D., de Carvalho, D. M. L., and Mareze, P. H.

(2020). Rendering binaural signals for moving sources. In Reproduced Sound

2020.

(Cited on page 20)

BIBLIOGRAPHY 204

[97] Grimm, G., Ewert, S., and Hohmann, V. (2015a). Evaluation of spatial

audio reproduction schemes for application in hearing aid research. Acta

Acustica united with Acustica, 101(4):842–854.

(Cited on pages 18, 21, 31, 41, 49, 53, 94, 117, 121, 160, 163, 177, 187, 190 and 191)

[98] Grimm, G., Kollmeier, B., and Hohmann, V. (2016a). Spatial Acoustic

Scenarios in Multichannel Loudspeaker Systems for Hearing Aid Evaluation.

Journal of the American Academy of Audiology, 27(7):557–566.

(Cited on pages 56, 163 and 191)

[99] Grimm, G., Kollmeier, B., and Hohmann, V. (2016b). Spatial Acoustic

Scenarios in Multichannel Loudspeaker Systems for Hearing Aid Evaluation.

Journal of the American Academy of Audiology.

(Cited on page 191)

[100] Grimm, G., Luberadzka, J., Herzke, T., and Hohmann, V. (2015b). Tool-

box for acoustic scene creation and rendering (TASCAR): Render methods

and research applications. Proceedings of the Linux Audio Conference.

(Cited on page 42)

[101] Grimm, G., Luberadzka, J., and Hohmann, V. (2018). Virtual acoustic

environments for comprehensive evaluation of model-based hearing devices

*. International Journal of Audiology.

(Cited on pages 42 and 120)

[102] Grimm, G., Luberadzka, J., and Hohmann, V. (2019). A toolbox for

rendering virtual acoustic environments in the context of audiology. Acta

Acustica united with Acustica, 105:566–578.

(Cited on pages 1, 42 and 57)

[103] Guastavino, C., Katz, B., Polack, J.-D., Levitin, D., and Dubois, D.

(2004). Ecological validity of soundscape reproduction. Acta Acustica united

with Acustica, 50.

(Cited on pages 42 and 191)

[104] Guastavino, C. and Katz, B. F. G. (2004). Perceptual evaluation of

multi-dimensional spatial audio reproduction. The Journal of the Acoustical

Society of America, 116:1105–1115.

(Cited on pages 57, 86 and 122)

[105] Guastavino, C., Larcher, V., Catusseau, G., and Boussard, P. (2007).

Spatial audio quality evaluation: comparing transaural, ambisonics and

stereo. In Proceedings of the 13th International Conference on Auditory

Display. Montr´eal Canada. Georgia Institute of Technology.

(Cited on pages 86 and 122)

[106] Hacihabiboglu, H., De Sena, E., Cvetkovic, Z., Johnston, J., and

Smith III, J. O. (2017). Perceptual spatial audio recording, simulation,

and rendering: An overview of spatial-audio techniques based on psychoa-

coustics. IEEE Signal Processing Magazine, 34(3):36–54.

(Cited on pages 20, 22 and 23)

[107] Hamdan, E. C. and Fletcher, M. D. (2022). A compact two-loudspeaker

virtual sound reproduction system for clinical testing of spatial hearing with

hearing-assistive devices. Frontiers in Neuroscience, 15.

(Cited on pages 42, 47 and 49)

BIBLIOGRAPHY 205

[108] Hammershøi, D. and Møller, H. (1992). Fundamentals of binaural tech-

nology. In Fundamentals of Binaural Technology.

(Cited on pages 12 and 16)

[109] Hammershøi, D. and Møller, H. (2005). Binaural technique — basic

methods for recording, synthesis, and reproduction. In Blauert, J., edi-

tor, Communication Acoustics, pages 223–254. Springer Berlin Heidelberg,

Berlin, Heidelberg.

(Cited on page 16)

[110] Harris, P., Nagy, S., and Vardaxis, N. (2018). Mosby’s Dictionary of

Medicine, Nursing and Health Professions - Revised 3rd Anz Edition. Else-

vier Health Sciences Apac.

(Cited on page 7)

[111] Havelock, D. I., Kuwano, S., and Vorlander, M. (2008). Handbook of

signal processing in acoustics. Springer, New York.

(Cited on page 98)

[112] Hazrati, O. and Loizou, P. C. (2012). The combined eﬀects of rever-

beration and noise on speech intelligibility by cochlear implant listeners.

International Journal of Audiology.

(Cited on page 98)

[113] He, J. (2016). Spatial Audio Reproduction with Primary Ambient Ex-

traction. SpringerBriefs in Electrical and Computer Engineering. Springer

Singapore.

(Cited on page 18)

[114] Hecker, S. (1984). Music for advertising eﬀect. Psychology & Marketing,

1(3-4):3–8.

(Cited on page 8)

[115] Hendrickx, E., Stitt, P., Messonnier, J.-C., Lyzwa, J.-M., Katz, B. F.,

and de Boish´eraud, C. (2017). Improvement of externalization by listener

and source movement using a “binauralized” microphone array. Journal of

the audio engineering society, 65(7/8):589–599.

(Cited on page 23)

[116] Hendrikse, M. M. E., Llorach, G., Hohmann, V., and Grimm, G. (2019).

Movement and gaze behavior in virtual audiovisual listening environments

resembling everyday life. Trends in Hearing, 23.

(Cited on pages 1 and 57)

[117] Hiyama, K., Komiyama, S., and Hamasaki, K. (2002). The minimum

number of loudspeakers and its arrangement for reproducing the spatial

impression of diﬀuse sound ﬁeld. Journal of the Audio Engineering Society.

(Cited on page 41)

[118] Hohmann, V., Paluch, R., Krueger, M., Meis, M., and Grimm, G. (2020).

The virtual reality lab: Realization and application of virtual sound envi-

ronments. Ear & Hearing, 41:31S–38S.

(Cited on pages 1 and 57)

BIBLIOGRAPHY 206

[119] Holman, J. A., Drummond, A., and Naylor, G. (2021). Hearing aids

reduce daily-life fatigue and increase social activity: a longitudinal study.

medRxiv.

(Cited on pages 1 and 56)

[120] Holube, I., Fredelake, S., Vlaming, M., and Kollmeier, B. (2010). Devel-

opment and analysis of an international speech test signal (ists). Interna-

tional Journal of Audiology, 49(12):891–903.

(Cited on page 131)

[121] Holube, I., Haeder, K., Imbery, C., and Weber, R. (2016). Subjective

Listening Eﬀort and Electrodermal Activity in Listening Situations with

Reverberation and Noise. Trends in hearing.

(Cited on pages 99 and 118)

[122] Hong, J. Y., He, J., Lam, B., Gupta, R., and Gan, W.-S. (2017). Spatial

audio for soundscape design: Recording and reproduction. Applied Sciences,

7(6).

(Cited on page 18)

[123] Hornsby, B. W. (2013). The eﬀects of hearing aid use on listening eﬀort

and mental fatigue associated with sustained speech processing demands.

Ear and hearing, 34(5):523–534.

(Cited on page 33)

[124] Howard, D. and Angus, J. (2009). Acoustics and Psychoacoustics 4th

Edition. Oxford: Focal Press, 4th edition.

(Cited on page 73)

[125] Huisman, T., Ahrens, A., and MacDonald, E. (2021). Ambisonics sound

source localization with varying amount of visual information in virtual

reality. Frontiers in Virtual Reality, 2.

(Cited on pages 16 and 49)

[126] International Telecommunications Union - Radiocommunications Sector

(ITU-R) (2015). Methods for the subjective assessment of small impair-

ments in audio systems. Technical report, International Telecommunications

Union, Geneva.

(Cited on pages 42, 61, 105 and 132)

[127] ISO (2009). 3382-1: Acoustics - measurement of room acoustic parame-

ters. part 1 : Performance spaces. ISO 1:2009, ISO.

(Cited on pages 33, 34, 35 and 70)

[128] J¨ancke, L. (2008). Music, memory and emotion. Journal of biology,

7(6):1–5.

(Cited on page 8)

[129] Jin, C., Corderoy, A., Carlile, S., and van Schaik, A. (2004). Contrasting

monaural and interaural spectral cues for human sound localization. The

Journal of the Acoustical Society of America, 115(6):3124–3141.

(Cited on page 11)

[130] Jot, J.-M., Wardle, S., and Larcher, V. (1998). approaches to binaural

synthesis. Journal of the Audio Engineering Society.

(Cited on page 27)

BIBLIOGRAPHY 207

[131] Kang, S. and Kim, S.-H. K. (1996). Realistic audio teleconferencing

using binaural and auralization techniques. Etri Journal, 18:41–51.

(Cited on page 23)

[132] Katz, B. F. G. and Noisternig, M. (2014). A comparative study of inter-

aural time delay estimation methods. The Journal of the Acoustical Society

of America, 135(6):3530–3540.

(Cited on page 70)

[133] Keet, V. (1968). The inﬂuence of early lateral reﬂections on the spatial

impression. Proc. 6th Int. Cong. Acoust., Tokyo, 2.

(Cited on page 39)

[134] Keidser, G., Naylor, G., Brungart, D. S., Caduﬀ, A., Campos, J., Carlile,

S., Carpenter, M. G., Grimm, G., Hohmann, V., Holube, I., Launer, S.,

Lunner, T., Mehra, R., Rapport, F., Slaney, M., and Smeds, K. (2020). The

quest for ecological validity in hearing science: what it is, why it matters,

and how to advance it. Ear and Hearing, 41(S1):5S–19S.

(Cited on pages 53, 56, 98, 122 and 191)

[135] Kestens, K., Degeest, S., and Keppler, H. (2021). The eﬀect of cognition

on the aided beneﬁt in terms of speech understanding and listening eﬀort

obtained with digital hearing aids: A systematic review. American Journal

of Audiology, 30(1):190–210.

(Cited on page 98)

[136] Kirsch, C., Poppitz, J., Wendt, T., van de Par, S., and Ewert, S. D.

(2021). Computationally eﬃcient spatial rendering of late reverberation

in virtual acoustic environments. 2021 Immersive and 3D Audio: from

Architecture to Automotive (I3DA).

(Cited on page 42)

[137] Klatte, M., Lachmann, T., Meis, M., et al. (2010). Eﬀects of noise and

reverberation on speech perception and listening comprehension of children

and adults in a classroom-like setting. Noise and Health, 12(49):270.

(Cited on pages 1 and 191)

[138] Kleiner, M., Dalenb¨ack, B.-I., and Svensson, P. (1993). Auralization-an

overview. Journal of the Audio Engineering Society, 41(11):861–875.

(Cited on page 19)

[139] Klemenz, M. (2005). Sound synthesis of starting electric railbound vehi-

cles and the inﬂuence of consonance on sound quality. Acta acustica united

with acustica, 91(4):779–788.

(Cited on page 20)

[140] Klockgether, S. and van de Par, S. (2016). Just noticeable diﬀerences of

spatial cues in echoic and anechoic acoustical environments. The Journal of

the Acoustical Society of America, 140(4):EL352–EL357.

(Cited on pages 93 and 94)

[141] Kobayashi, M., Ueno, K., and Ise, S. (2015). The Eﬀects of Spatial-

ized Sounds on the Sense of Presence in Auditory Virtual Environments: A

Psychological and Physiological Study. Presence: Teleoperators and Virtual

Environments, 24(2):163–174.

(Cited on page 16)

BIBLIOGRAPHY 208

[142] Koehnke, J. and Besing, J. (1996). A procedure for testing speech intelli-

gibility in a virtual listening environment. Ear and Hearing, 17(3):211–217.

cited By 59.

(Cited on page 40)

[143] Koelewijn, T., Zekveld, A. A., Festen, J. M., and Kramer, S. E. (2012).

Pupil dilation uncovers extra listening eﬀort in the presence of a single-talker

masker. Ear and Hearing, 33(2):291–300.

(Cited on page 190)

[144] Kramer, S. E., Bhuiyan, T., Bramsløw, L., Fiedler, L., Graversen, C.,

Hadley, L. V., Innes-Brown, H., Naylor, G., Richter, M., Saunders, G. H.,

Versfeld, N. J., Wendt, D., Whitmer, W. M., and Zekveld, A. A. (2020).

Innovative hearing aid research on ecological conditions and outcome mea-

sures: The hear-eco project.

(Cited on page 191)

[145] Kramer, S. E., Kapteyn, T. S., Festen, J. M., and Tobi, H. (1996).

The relationships between self-reported hearing disability and measures of

auditory disability. Audiology, 35(5):277–287.

(Cited on page 56)

[146] Krokstad, A., Strom, S., and Sørsdal, S. (1968). Calculating the acous-

tical room response by the use of a ray tracing technique. Journal of Sound

and Vibration, 8(1):118–125.

(Cited on page 19)

[147] Krueger, M., Schulte, M., Brand, T., and Holube, I. (2017). Development

of an adaptive scaling method for subjective listening eﬀort. The Journal of

the Acoustical Society of America.

(Cited on page 50)

[148] Kuttruﬀ, H. (2009). Room Acoustics, Fifth Edition. Taylor & Francis.

(Cited on pages 19, 34, 68 and 127)

[149] Kwak, C., Han, W., Lee, J., Kim, J., and Kim, S. (2018). Eﬀect of noise

and reverberation on speech recognition and listening eﬀort for older adults.

Geriatrics and Gerontology International.

(Cited on pages 50, 99 and 118)

[150] Laitinen, M.-V. and Pulkki, V. (2009). Binaural reproduction for di-

rectional audio coding. In 2009 IEEE Workshop on Applications of Signal

Processing to Audio and Acoustics, pages 337–340.

(Cited on page 41)

[151] Lau, M. K., Hicks, C., Kroll, T., and Zupancic, S. (2019). Eﬀect of

auditory task type on physiological and subjective measures of listening

eﬀort in individuals with normal hearing. Journal of Speech, Language, and

Hearing Research.

(Cited on pages 50 and 52)

[152] Lau, S.-T., Pichora-Fuller, M., Li, K., Singh, G., and Campos, J. (2016).

Eﬀects of hearing loss on dual-task performance in an audiovisual virtual re-

ality simulation of listening while walking. Journal of the American Academy

of Audiology, 27.

(Cited on page 57)

BIBLIOGRAPHY 209

[153] Letowski, T. and Letowski, S. (2011). Localization error accuracy and

precision of auditory localization. In Strumillo, P., editor, Advances in Sound

Localization, chapter 4, pages 55–78. Intech, Oxford.

(Cited on pages 9, 10 and 17)

[154] Levy, S. M. (2012). Section 9 - calculations to determine the eﬀec-

tiveness and control of thermal and sound transmission. In Levy, S. M.,

editor, Construction Calculations Manual, pages 503–544. Butterworth-

Heinemann, Boston.

(Cited on page 60)

[155] Lindau, A. and Brinkmann, F. (2012). perceptual evaluation of head-

phone compensation in binaural synthesis based on non-individual record-

ings. journal of the audio engineering society, 60(1/2):54–62.

(Cited on page 14)

[156] Lindau, A., Kosanke, L., and Weinzierl, S. (2010). perceptual evaluation

of physical predictors of the mixing time in binaural room impulse responses.

Journal of the Audio Engineering Society.

(Cited on page 126)

[157] Lindemann, W. (1986). Extension of a binaural cross-correlation model

by contralateral inhibition. i. simulation of lateralization for stationary sig-

nals. The Journal of the Acoustical Society of America, 80 6:1608–22.

(Cited on page 45)

[158] Liu, Z., Fard, M., and Jazar, R. (2015). Development of an acoustic

material database for vehicle interior trims. Technical report, SAE Technical

Paper.

(Cited on page 21)

[159] Llopis, H. S., Pind, F., and Jeong, C.-H. (2020). Development of

an auditory virtual reality system based on pre-computed b-format im-

pulse responses for building design evaluation. Building and Environment,

169:106553.

(Cited on page 133)

[160] Llorach, G., Evans, A., Blat, J., Grimm, G., and Hohmann, V. (2016).

Web-based live speech-driven lip-sync. In 2016 8th International Conference

on Games and Virtual Worlds for Serious Applications (VS-GAMES), pages

1–4.

(Cited on pages 1 and 57)

[161] Llorach, G., Grimm, G., Hendrikse, M. M., and Hohmann, V. (2018).

Towards Realistic Immersive Audiovisual Simulations for Hearing Research.

In Proceedings of the 2018 Workshop on Audio-Visual Scene Understanding

for Immersive Multimedia, pages 33–40.

(Cited on pages 1, 18, 27, 40, 53, 56, 57, 58 and 191)

[162] Llorca-Bof´ı, J., Dreier, C., Heck, J., and Vorl¨ander, M. (2022). Urban

sound auralization and visualization framework;case study at ihtapark. Sus-

tainability, 14(4).

(Cited on page 20)

BIBLIOGRAPHY 210

[163] Loizou, P. C. (2007). Speech enhancement: theory and practice. CRC

press.

(Cited on page 52)

[164] Lokki, T. and Savioja, L. (2008). Virtual acoustics. In Havelock, D.,

Kuwano, S., and Vorl¨ander, M., editors, Handbook of Signal Processing in

Acoustics, pages 761–771. Springer New York, New York, NY.

(Cited on page 20)

[165] Long, M. (2014). Architectural Acoustics. Elsevier Science.

(Cited on pages 16, 19 and 42)

[166] Lopez, J. J., Gutierrez, P., Cobos, M., and Aguilera, E. (2014). Sound

distance perception comparison between Wave Field Synthesis and Vector

Base Amplitude Panning. In ISCCSP 2014 - 2014 6th International Sym-

posium on Communications, Control and Signal Processing, Proceedings.

(Cited on pages 21 and 121)

[167] Lovedee-Turner, M. and Murphy, D. (2018). Application of machine

learning for the spatial analysis of binaural room impulse responses. Applied

Sciences, 8(1).

(Cited on page 15)

[168] Lund, K. D., Ahrens, A., and Dau, T. (2020). A method for evalu-

ating audio-visual scene analysis in multi-talker environments. In Proceed-

ings of the International Symposium on Auditory and Audiological Research,

volume 7, pages 357–364. The Danavox Jubilee Foundation. International

Symposium on Auditory and Audiological Research ISAAR2019.

(Cited on page 42)

[169] Lundbeck, M., Grimm, G., Hohmann, V., Laugesen, S., and Neher, T.

(2017). Sensitivity to Angular and Radial Source Movements as a Function

of Acoustic Complexity in Normal and Impaired Hearing. Trends in Hearing,

21:2331–2165.

(Cited on pages 33 and 57)

[170] Lyon, R. F. (2017). Human and Machine Hearing Extracting Meaning

from Sound. a. Cambridge University Press.

(Cited on page 10)

[171] Magezi, D. A. (2015). Linear mixed-eﬀects models for within-participant

psychology experiments: an introductory tutorial and free, graphical user

interface (lmmgui). Frontiers in Psychology, 6:2.

(Cited on page 113)

[172] Malham, D. G. and Myatt, A. (1995). 3-d sound spatialization using

ambisonic techniques. Computer Music Journal, 19(4):58–70.

(Cited on page 29)

[173] Mansour, N., Marschall, M., May, T., Westermann, A., and Dau, T.

(2021a). Speech intelligibility in a realistic virtual sound environment. The

Journal of the Acoustical Society of America, 149(4):2791–2801.

(Cited on page 99)

BIBLIOGRAPHY 211

[174] Mansour, N., Westermann, A., Marschall, M., May, T., Dau, T., and

Buchholz, J. (2021b). Guided ecological momentary assessment in real

and virtual sound environments. Acoustical Society of America. Journal,

150(4):2695–2704.

(Cited on pages 16 and 42)

[175] Marentakis, G., Zotter, F., and Frank, M. (2014). Vector-base and am-

bisonic amplitude panning: A comparison using pop, classical, and contem-

porary spatial music. Acta Acustica united with Acustica.

(Cited on pages 57 and 86)

[176] Marrone, N., Mason, C. R., and Kidd, G. (2008). The eﬀects of hearing

loss and age on the beneﬁt of spatial separation between multiple talkers

in reverberant rooms. The Journal of the Acoustical Society of America,

124(5):3064–3075.

(Cited on pages 16 and 40)

[177] Marschall, M. (2014). Capturing and reproducing realistic acoustic

scenes for hearing research. PhD Thesis - Technical University of Denmark.

(Cited on pages 40, 53, 99 and 120)

[178] Masiero, B. (2012). Individualized Binaural Technology. Measurement,

Equalization and Perceptual Evaluation. PhD thesis, RWTH Aachen Uni-

versity.

(Cited on page 14)

[179] Masiero, B. and Fels, J. (2011). Perceptually robust headphone equal-

ization for binaural reproduction. In Audio Engineering Society Convention

130. Audio Engineering Society.

(Cited on page 22)

[180] Masiero, B. and Vorlaender, M. (2011). Spatial Audio Reproduction

Methods for Virtual Reality. In 42º Congreso Espa˜nol de Ac´ustica Encuentro

Ib´erico de Ac´ustica - European Symposium on Environmental Acoustics and

on Buildings Acoustically Sustainable, pages 1–12, C´aceres.

(Cited on pages 23, 24, 58 and 86)

[181] Matthen, M. (2016). Eﬀort and displeasure in people who are hard of

hearing. Ear and Hearing, 37 Suppl 1.

(Cited on page 57)

[182] May, T., van de Par, S., and Kohlrausch, A. (2011). A probabilistic

model for robust localization based on a binaural auditory front-end. IEEE

Transactions on Audio, Speech, and Language Processing, 19(1):1–13.

(Cited on pages xix, 6, 138, 150, 156, 166, 167, 174 and 178)

[183] Meesawat, K. and Hammershoi, D. (2003). The time when the reverber-

ation tail in a binaural room impulse response begins. In Audio Engineering

Society Convention 115. Audio Engineering Society.

(Cited on page 15)

[184] Menase, D. A., Richter, M., Wendt, D., Fiedler, L., and Naylor, G.

(2022). Task-induced mental fatigue and motivation inﬂuence listening eﬀort

as measured by the pupil dilation in a speech-in-noise task. medRxiv.

(Cited on page 33)

BIBLIOGRAPHY 212

[185] Michael, V. and Vorl¨ander, M. (2008). Auralization. Fundamentals of

Acoustics, Modelling, Simulation, Algorithms and Acoustic Virtual Reality.

Springer.

(Cited on page 98)

[186] Miles, K., McMahon, C., Boisvert, I., Ibrahim, R., de Lissa, P., Gra-

ham, P., and Lyxell, B. (2017). Objective Assessment of Listening Eﬀort:

Coregistration of Pupillometry and EEG. Trends in Hearing.

(Cited on pages 50, 51 and 118)

[187] Millington, G. (1932). A modiﬁed formula for reverberation. The Journal

of the Acoustical Society of America, 4(1A):69–82.

(Cited on page 34)

[188] Minnaar, P., Favrot, S., and Buchholz, J. (2010). Improving hearing aids

through listening tests in a virtual sound environment. Hearing Journal,

63(10):40–44.

(Cited on pages 1, 16, 42, 121 and 191)

[189] Møller, H., Sørensen, M. F., Hammershøi, D., and Jensen, C. B. (1995).

Head-related transfer functions of human subjects. Journal of the Audio

Engineering Society, 43(5):300–321.

(Cited on page 12)

[190] Monaghan, J. J., Krumbholz, K., and Seeber, B. U. (2013). Factors

aﬀecting the use of envelope interaural time diﬀerences in reverberationa).

The Journal of the Acoustical Society of America, 133(4):2288–2300. bibtex:

Monaghan2013.

(Cited on page 33)

[191] Moore, B. C. J. and Tan, C.-T. (2004). development and validation of

a method for predicting the perceived naturalness of sounds subjected to

spectral distortion. Journal of the Audio Engineering Society, 52(9):900–

914.

(Cited on page 41)

[192] Moore, T. M. and Picou, E. M. (2018). A potential bias in subjective

ratings of mental eﬀort. Journal of Speech, Language, and Hearing Research.

(Cited on pages 50, 51 and 118)

[193] Mueller, M. F., Kegel, A., Schimmel, S. M., Dillier, N., and Hofbauer, M.

(2012). Localization of virtual sound sources with bilateral hearing aids in

realistic acoustical scenes. The Journal of the Acoustical Society of America,

131(6):4732–4742.

(Cited on page 16)

[194] M¨uller, S. and Massarani, P. (2001). Transfer-function measurement

with sweeps. Journal of the Audio Engineering Society, 49:443–471.

(Cited on pages 63 and 140)

[195] Murta, B. (2019). Plataforma para ensaios de percep¸c˜ao sonora com

fontes distribu´ıdas aplic´avel a dispositivos auditivos: perSONA (in Por-

tuguese). PhD thesis, Federal University of Santa Catarina.

(Cited on pages 1 and 57)

BIBLIOGRAPHY 213

[196] Murta, B., Chiea, R., Mour˜ao, G., Pinheiro, M. M., Cordioli, J., Paul, S.,

and Costa, M. (2019). Cci-mobile: Development of software based tools for

speech perception assessment and training with hearing impaired brazil-

ian population. In CONFERENCE on Implantable Auditory Prostheses

(CIAP), Lake Tahoe, California, US.

(Cited on page 18)

[197] Møller, H. (1992). Fundamentals of binaural technology. Applied Acous-

tics, 36(3-4):171–218.

(Cited on pages 15 and 73)

[198] Nachbar, C., Zotter, F., Deleﬂie, E., and Sontacchi, A. (2011). Ambix –

a suggested ambisonics format.

(Cited on page 103)

[199] Narbutt, M., Allen, A., Skoglund, J., Chinen, M., and Hines, A. (2018).

Ambiqual - a full reference objective quality metric for ambisonic spatial

audio. In 2018 Tenth International Conference on Quality of Multimedia

Experience (QoMEX), pages 1–6.

(Cited on page 27)

[200] Naugolnykh, K. A., Ostrovsky, L. A., Sapozhnikov, O. A., and Hamilton,

M. F. (2000). Nonlinear wave processes in acoustics.

(Cited on page 9)

[201] Naylor, G. M. (1993). Odeon—another hybrid room acoustical model.

Applied Acoustics, 38(2-4):131–143.

(Cited on page 68)

[202] Neuhoﬀ, J. (2021). Ecological psychoacoustics. Brill.

(Cited on page 190)

[203] Neuman, A. C., Wroblewski, M., Hajicek, J., and Rubinstein, A. (2010).

Combined eﬀects of noise and reverberation on speech recognition perfor-

mance of normal-hearing children and adults. Ear and Hearing.

(Cited on pages 99 and 118)

[204] Nicola, P. and Chiara, V. (2019). Impact of Background Noise Fluc-

tuation and Reverberation on Response Time in a Speech Reception Task.

Journal of Speech, Language, and Hearing Research, 62(11):4179–4195.

(Cited on pages 50, 99 and 118)

[205] Nielsen, J. and Dau, T. (2011). The danish hearing in noise test. Inter-

national journal of audiology, 50:202–8.

(Cited on pages 101 and 102)

[206] Nocke, C. and Mellert, V. (2002). Brief review on in situ measurement

techniques of impedance or absorption. In Forum Acusticum, Sevilla.

(Cited on page 21)

[207] Novo, P. (2005). Auditory virtual environments. In Blauert, J., edi-

tor, Communication Acoustics, pages 277–297. Springer Berlin Heidelberg,

Berlin, Heidelberg.

(Cited on page 57)

BIBLIOGRAPHY 214

[208] Obleser, J., W¨ostmann, M., Hellbernd, N., Wilsch, A., and Maess, B.

(2012). Adverse listening conditions and memory load drive a common alpha

oscillatory network. Journal of Neuroscience, 32(36):12376–12383.

(Cited on page 111)

[209] Ohlenforst, B., Wendt, D., Kramer, S. E., Naylor, G., Zekveld, A. A.,

and Lunner, T. (2018). Impact of SNR, masker type and noise reduction

processing on sentence recognition performance and listening eﬀort as indi-

cated by the pupil dilation response. Hearing Research.

(Cited on pages 50 and 101)

[210] Ohlenforst, B., Zekveld, A. A., Jansma, E. P., Wang, Y., Naylor, G.,

Lorens, A., Lunner, T., and Kramer, S. E. (2017a). Eﬀects of hearing

impairment and hearing aid ampliﬁcation on listening eﬀort: A systematic

review. Ear and hearing, 38(3):267—281.

(Cited on page 98)

[211] Ohlenforst, B., Zekveld, A. A., Lunner, T., Wendt, D., Naylor, G., Wang,

Y., Versfeld, N. J., and Kramer, S. E. (2017b). Impact of stimulus-related

factors and hearing impairment on listening eﬀort as indicated by pupil

dilation. Hearing Research, 351:68–79.

(Cited on page 50)

[212] Oreinos, C. and Buchholz, J. (2014). Validation of realistic acoustic

environments for listening tests using directional hearing aids. In 2014 14th

International Workshop on Acoustic Signal Enhancement (IWAENC), pages

188–192.

(Cited on pages 41 and 120)

[213] Oreinos, C. and Buchholz, J. M. (2015). Objective analysis of ambisonics

for hearing aid applications: Eﬀect of listener’s head, room reverberation,

and directional microphones. The Journal of the Acoustical Society of Amer-

ica.

(Cited on pages 18, 41, 53, 163 and 191)

[214] Palacino, J., Nicol, R., Emerit, M., and Gros, L. (2012). Perceptual

assessment of binaural decoding of ﬁrst-order ambisonics. In Acoustics 2012.

(Cited on page 21)

[215] Parsehian, G., Gandemer, L., Bourdin, C., and Kronland Martinet,

R. (2015). Design and perceptual evaluation of a fully immersive three-

dimensional sound spatialization system. In 3rd International Conference

on Spatial Audio (ICSA 2015), Graz, Austria.

(Cited on page 42)

[216] Paul, S. (13-15 maio 2014). A ﬁsiologia da audi¸c˜ao como base para

fenˆomenos auditivos. In Proceedings of the 12th AES Brazil Conference,

S˜ao Paulo, SP.

(Cited on page 9)

[217] Pausch, F., Asp¨ock, L., Vorl¨ander, M., and Fels, J. (2018). An Ex-

tended Binaural Real-Time Auralization System With an Interface to Re-

search Hearing Aids for Experiments on Subjects With Hearing Loss. Trends

in Hearing.

(Cited on pages 16, 44, 45, 120 and 121)

BIBLIOGRAPHY 215

[218] Pausch, F., Behler, G., and Fels, J. (2020). Scalar - a surrounding spher-

ical cap loudspeaker array for ﬂexible generation and evaluation of virtual

acoustic environments. Acta Acust., 4(5):19.

(Cited on pages 1 and 57)

[219] Pausch, F. and Fels, J. (2019). Mobilab – a mobile laboratory for on-site

listening experiments in virtual acoustic environments. bioRxiv.

(Cited on pages 1 and 57)

[220] Pausch, F. and Fels, J. (2020). Localization performance in a binaural

real-time auralization system extended to research hearing aids. Trends in

hearing, 24:1–18.

(Cited on pages 1, 42 and 57)

[221] Pelzer, S., Masiero, B., and Vorl¨ander, M. (2014). 3D Reproduction of

Room Auralizations by Combining Intensity Panning, Crosstalk Cancella-

tion and Ambisonics. Proceedings of the EAA Joint Symposium on Aural-

ization and Ambisonics.

(Cited on pages 44, 45, 86 and 127)

[222] Peng, Z. E. and Litovsky, R. Y. (2021). The role of interaural diﬀerences,

head shadow, and binaural redundancy in binaural intelligibility beneﬁts

among school-aged children. Trends in Hearing, 25.

(Cited on page 77)

[223] Petersen, E. B., W¨ostmann, M., Obleser, J., Stenfelt, S., and Lunner,

T. (2015). Hearing loss impacts neural alpha oscillations under adverse

listening conditions. Frontiers in Psychology.

(Cited on page 50)

[224] Pichora-Fuller, M. K., Kramer, S. E., Eckert, M. A., Edwards, B.,

Hornsby, B. W., Humes, L. E., Lemke, U., Lunner, T., Matthen, M., Mack-

ersie, C. L., Naylor, G., Phillips, N. A., Richter, M., Rudner, M., Sommers,

M. S., Tremblay, K. L., and Wingﬁeld, A. (2016). Hearing impairment

and cognitive energy: The framework for understanding eﬀortful listening

(FUEL). In Ear and Hearing.

(Cited on pages 50, 51, 52, 57 and 118)

[225] Picou, E. M., Gordon, J., and Ricketts, T. A. (2016). The eﬀects of noise

and reverberation on listening eﬀort in adults with normal hearing. Ear and

Hearing.

(Cited on pages 50 and 99)

[226] Picou, E. M., Moore, T. M., and Ricketts, T. A. (2017). The eﬀects of

directional processing on objective and subjective listening eﬀort. Journal

of Speech, Language, and Hearing Research.

(Cited on pages 1 and 51)

[227] Picou, E. M., Ricketts, T., and Hornsby, B. (2013). How hearing aids,

background noise, and visual cues inﬂuence objective listening eﬀort. Ear

and Hearing, 34:e52–e64.

(Cited on pages 50 and 56)

BIBLIOGRAPHY 216

[228] Picou, E. M. and Ricketts, T. A. (2014). Increasing motivation changes

subjective reports of listening eﬀort and choice of coping strategy. Interna-

tional Journal of Audiology, 53(6):418–426.

(Cited on page 50)

[229] Picou, E. M. and Ricketts, T. A. (2018). The relationship between speech

recognition, behavioural listening eﬀort, and subjective ratings. Interna-

tional Journal of Audiology.

(Cited on pages 51 and 118)

[230] Pielage, H., Zekveld, A. A., Saunders, G. H., Versfeld, N. J., Lunner, T.,

and Kramer, S. E. (2021). The Presence of Another Individual Inﬂuences

Listening Eﬀort, But Not Performance. Ear & Hearing.

(Cited on pages 40, 57, 82 and 190)

[231] Pieren, R. (2018). Auralization of Environmental Acoustical Sceneries:

Synthesis of Road Traﬃc, Railway and Wind Turbine Noise. PhD thesis,

Delft University of Technology.

(Cited on page 20)

[232] Pieren, R., Heutschi, K., Wunderli, J. M., Snellen, M., and Simons,

D. G. (2017). Auralization of railway noise: Emission synthesis of rolling

and impact noise. Applied Acoustics, 127:34–45.

(Cited on page 20)

[233] Pinheiro, J. C. and Bates, D. M. (2000). Linear mixed-eﬀects models:

basic concepts and examples. Mixed-eﬀects models in S and S-Plus, pages

3–56.

(Cited on page 113)

[234] Plain, B., Pielage, H., Richter, M., Bhuiyan, T., Lunner, T., Kramer, S.,

and Zekveld, A. (2021). Social observation increases the cardiovascular re-

sponse of hearing-impaired listeners during a speech reception task. Hearing

Research, page 108334.

(Cited on pages 57 and 190)

[235] Plinge, A., Schlecht, S. J., Thiergart, O., Robotham, T., Rummukainen,

O., and Habets, E. A. P. (2018). six-degrees-of-freedom binaural audio

reproduction of ﬁrst-order ambisonics with distance information. Journal of

the audio engineering society.

(Cited on page 27)

[236] Poletti, M. A. (2005). Three-dimensional surround sound systems

based on spherical harmonics. journal of the audio engineering society,

53(11):1004–1025.

(Cited on page 31)

[237] Politis, A. (2016). Microphone array processing for parametric spatial

audio techniques. Doctoral thesis, School of Electrical Engineering.

(Cited on pages 130, 131 and 181)

[238] Politis, A., McCormack, L., and Pulkki, V. (2017). Enhancement of

ambisonic binaural reproduction using directional audio coding with opti-

mal adaptive mixing. In 2017 IEEE Workshop on Applications of Signal

Processing to Audio and Acoustics (WASPAA), pages 379–383.

(Cited on page 27)

BIBLIOGRAPHY 217

[239] Pollow, M. (2015). Directivity Patterns for Room Acoustical Measure-

ments and Simulations. Aachener Beitr¨age zur Technischen Akustik. Logos

Verlag Berlin GmbH.

(Cited on pages 28 and 29)

[240] Portela, M. S. (2008). Caracteriza¸c˜ao de fontes sonoras e aplica¸c˜ao na

auraliza¸c˜ao de ambientes. Mestrado, Universidade Federal de Santa Cata-

rina.

(Cited on page 13)

[241] Pulkki, V. (1997). Virtual sound source positioning using vector base

amplitude panning. Journal of the Audio Engineering Society, 45(6).

(Cited on pages 23, 24, 26, 58, 93, 142, 179 and 190)

[242] Pulkki, V. and Karjalainen, M. (2015). Communication Acoustics: An

Introduction to Speech, Audio and Psychoacoustics. Wiley.

(Cited on pages 10, 17, 70, 73 and 143)

[243] Pulkki, V., Politis, A., Laitinen, M.-V., Vilkamo, J., and Ahonen, J.

(2017). First-order directional audio coding (dirac). In Parametric Time-

Frequency Domain Spatial Audio, chapter 5, pages 89–140. John Wiley &

Sons, Ltd.

(Cited on pages 44, 45 and 127)

[244] Purdy, M. (1991). Listening and community: The role of listening in

community formation. International Journal of Listening, 5(1):51–67.

(Cited on page 8)

[245] Queiroz, M., Iazzetta, F., Kon, F., Gomes, M. H. A., Figueiredo, F. L.,

Masiero, B. S., Dias, L., Torres, M. H. C., and Thomaz, L. F. (2008). Acmus:

an open, integrated platform for room acoustics research. J. Braz. Comput.

Soc., 14(3):87–103.

(Cited on page 36)

[246] Rayleigh, L. (1907). Xii. on our perception of sound direction.

(Cited on page 10)

[247] Reichardt, W., Alim, O. A., and Schmidt, W. (1975). Deﬁnition and

basis of making an objective evaluation to distinguish between useful and

useless clarity deﬁning musical performances. Acta Acustica united with

Acustica, 32(3):126–137.

(Cited on page 36)

[248] Rennies, J., Brand, T., and Kollmeier, B. (2011). Prediction of the

inﬂuence of reverberation on binaural speech intelligibility in noise and in

quiet. The Journal of the Acoustical Society of America, 130(5):2999–3012.

(Cited on page 40)

[249] Rennies, J., Schepker, H., Holube, I., and Kollmeier, B. (2014). Listening

eﬀort and speech intelligibility in listening situations aﬀected by noise and

reverberation. The Journal of the Acoustical Society of America.

(Cited on page 50)

BIBLIOGRAPHY 218

[250] Roﬄer, S. K. and Butler, R. A. (1968). Factors that inﬂuence the local-

ization of sound in the vertical plane. The Journal of the Acoustical Society

of America, 43(6):1255–1259.

(Cited on pages 10 and 33)

[251] Roginska, A. (2017). Binaural audio through headphones. In Immersive

Sound, pages 88–123. Routledge.

(Cited on pages 40 and 53)

[252] Romanov, M., Berghold, P., Frank, M., Rudrich, D., Zaunschirm, M.,

and Zotter, F. (2017). Implementation and evaluation of a low-cost head-

tracker for binaural synthesis. Journal of the audio engineering society.

(Cited on page 23)

[253] Rose, J., Nelson, P., Rafaely, B., and Takeuchi, T. (2002). Sweet spot

size of virtual acoustic imaging systems at asymmetric listener locations.

The Journal of the Acoustical Society of America, 112(5):1992–2002.

(Cited on pages 31 and 121)

[254] Rossing, T. D. (2007). Springer Handbook of Acoustics. Springer Hand-

book of Acoustics. Springer-Verlag Berlin Heidelberg, Stanford, CA, 2 edi-

tion.

(Cited on pages 16, 19, 33, 36, 38 and 98)

[255] Rudenko, O. and Soluian, S. (1975). The theoretical principles of non-

linear acoustics. Moscow Izdatel Nauka.

(Cited on page 9)

[256] Rumsey, F. (2013). Spatial Audio. Focal Press, Burlington, MA, 2 edi-

tion.

(Cited on pages 30 and 98)

[257] Ruotolo, F., Maﬀei, L., Di Gabriele, M., Iachini, T., Masullo, M., Rug-

giero, G., and Senese, V. P. (2013). Immersive virtual reality and environ-

mental noise assessment: An innovative audio–visual approach. Environ-

mental Impact Assessment Review, 41:10–20.

(Cited on page 16)

[258] Sabine, W. (1922). Collected Papers on Acoustics. Harvard University

Press.

(Cited on page 34)

[259] Savioja, L., Huopaniemi, J., Lokki, T., and Vaananen, R. (1999). Cre-

ating interactive virtual acoustic environments. Journal of the Audio Engi-

neering Society, 47:675–705.

(Cited on pages 1 and 57)

[260] Schepker, H., Haeder, K., Rennies, J., and Holube, I. (2016). Per-

ceived listening eﬀort and speech intelligibility in reverberation and noise

for hearing-impaired listeners. International Journal of Audiology.

(Cited on pages 1 and 50)

[261] Schr¨oder, D. (2011). Physically Based Real-Time Auralization of Inter-

active Virtual Environments. Aachener Beitr¨age zur Technischen Akustik.

Logos Verlag Berlin.

(Cited on page 28)

BIBLIOGRAPHY 219

[262] Schroeder, M. and Atal, B. (1963). Computer simulation of sound trans-

mission in rooms. Proceedings of the IEEE, 51(3):536–537.

(Cited on page 22)

[263] Schroeder, M., Atal, B., and Bird, C. (1962). Digital computers in room

acoustics. Proc. 4th ICA, Copenhagen M, 21.

(Cited on page 19)

[264] Schroeder, M. R. (1965). New method of measuring reverberation time.

The Journal of the Acoustical Society of America, 37(3):409–412.

(Cited on page 144)

[265] Schroeder, M. R. (1979). Integrated impulse method measuring sound

decay without using impulses. The Journal of the Acoustical Society of

America, 66(2):497–500.

(Cited on page 144)

[266] Schr¨oder, D., Pohl, A., Drechsler, S., Svensson, U. P., Vorl¨ander, M.,

and Stephenson, U. M. (2013). openmat - management of acoustic material

(meta-)properties using an open source database format. In Proceedings of

the AIA-DAGA 2013.

(Cited on page 21)

[267] Schr¨oder, D., Wefers, F., Pelzer, S., Rausch, D., Vorlaender, M., and

Kuhlen, T. (2010). Virtual reality system at rwth aachen university. In

Proceedings of the International Symposium on Room Acoustics (ISRA).

(Cited on page 56)

[268] Seeber, B. U., Baumann, U., and Fastl, H. (2004). Localization ability

with bimodal hearing aids and bilateral cochlear implants. The Journal of

the Acoustical Society of America, 116(3):1698–1709.

(Cited on page 40)

[269] Seeber, B. U., Kerber, S., and Hafter, E. R. (2010). A system to simu-

late and reproduce audio–visual environments for spatial hearing research.

Hearing research, 260(1):1–10.

(Cited on page 56)

[270] Seikel, J., King, D., and Drumright, D. (2015). Anatomy & Physiology

for Speech, Language, and Hearing. Cengage Learning.

(Cited on page 14)

[271] Sette, W. J. (1933). A new reverberation time formula. The Journal of

the Acoustical Society of America, 4(3):193–210.

(Cited on page 34)

[272] Shavit-Cohen, K. and Zion Golumbic, E. (2019). The dynamics of atten-

tion shifts among concurrent speech in a naturalistic multi-speaker virtual

environment. Frontiers in Human Neuroscience, 13:386.

(Cited on pages 1 and 57)

[273] Shojaei, E., Ashayeri, H., Jafari, Z., Dast, M., and Kamali, K. (2016).

Eﬀect of signal to noise ratio on the speech perception ability of older adults.

Medical journal of the Islamic Republic of Iran, 30:342.

(Cited on pages 1 and 55)

BIBLIOGRAPHY 220

[274] Silzle, A., Kosmidis, D., Felix Greco, G., Beer, D., and Betz, L. (2016).

The inﬂuence of microphone directivity on the level calibration and equal-

ization of 3d loudspeakers setups. In 29th Tonmeistertagung - VDT Inter-

national Convention 2016.

(Cited on page 21)

[275] Simon, L. S. R., Dillier, N., and W¨uthrich, H. (2021). Comparison of

3D audio reproduction methods using hearing devices. Journal of the Audio

Engineering Society, 68(12):899–909.

(Cited on pages 21, 46, 93, 94 and 121)

[276] Simon, L. S. R., Wuethrich, H., and Dillier, N. (2017). Comparison

of higher-order ambisonics, vector- and distance-based amplitude panning

using a hearing device beamformer. In Proceedings of 4th International

Conference on Spatial Audio, Graz, Austria.

(Cited on pages 20, 21, 23, 117, 121, 163 and 191)

[277] Sim´on G´alvez, M., Menzies, D., Fazi, F., de Campos, T., and Hilton,

A. (2015). Listener tracking stereo for object based audio reproduction.

In Tecniacustica 2016 (Valencia)-European Symposium in Virtual Acoustics

and Ambisonics.

(Cited on page 27)

[278] Skudrzyk, E. (1971). The foundations of acoustics: basic mathematics

and basic acoustics. Springer-Verlag.

(Cited on page 229)

[279] Solvang, A. (2008). Spectral impairment of two-dimensional higher order

ambisonics. J. Audio Eng. Soc, 56(4):267–279.

(Cited on page 94)

[280] Spand¨ock, F. (1934). Akustische modellversuche. Annalen der Physik,

412(4):345–360.

(Cited on page 19)

[281] Spors, S., Teutsch, H., Kuntz, A., and Rabenstein, R. (2004). Sound ﬁeld

synthesis. In Huang, Y. and Benesty, J., editors, Audio Signal Processing

for Next-Generation Multimedia Communication Systems, pages 323–344.

Springer US, Boston, MA.

(Cited on page 31)

[282] Spors, S., Wierstorf, H., Raake, A., Melchior, F., Frank, M., and Zotter,

F. (2013). Spatial sound with loudspeakers and its perception: A review of

the current state.

(Cited on pages 21, 23, 27, 40, 42 and 53)

[283] Stitt, P., Bertet, S., and Van Walstijn, M. (2013). Perceptual investiga-

tion of image placement with ambisonics for non-centred listeners. In Proc.

of the 16th Int. Conference on Digital Audio Eﬀects (DAFx-13), Maynooth,

Ireland.

(Cited on pages 21, 46 and 49)

[284] Strauss, H. (1998). Implementing doppler shifts for virtual auditory

environments. Journal of the Audio Engineering Society.

(Cited on page 20)

BIBLIOGRAPHY 221

[285] Strumi l lo, P. (2011). Advances in Sound Localization. a. InTech.

(Cited on pages 9 and 10)

[286] Sudarsono, A. S., Lam, Y. W., and Davies, W. J. (2016). The eﬀect

of sound level on perception of reproduced soundscapes. Applied Acoustics,

110:53–60.

(Cited on page 42)

[287] Søndergaard, P. and Majdak, P. (2013). The auditory modeling toolbox.

In Blauert, J., editor, The Technology of Binaural Listening, pages 33–56.

Springer, Berlin, Heidelberg.

(Cited on page 138)

[288] Tenenbaum, R. A., Camilo, T. S., Torres, J. C. B., and Gerges, S. N.

(2007). Hybrid method for numerical simulation of room acoustics with au-

ralization: part 1-theoretical and numerical aspects. Journal of the Brazilian

Society of Mechanical Sciences and Engineering, 29:211–221.

(Cited on page 68)

[289] Tremblay, P., Brisson, V., and Deschamps, I. (2020). Brain aging and

speech perception: Eﬀects of background noise and talker variability. Neu-

roImage, 227:117675.

(Cited on pages 1 and 55)

[290] Trevi˜no, J., Okamoto, T., Iwaya, Y., and Suzuki, Y. (2011). Evaluation

of a new ambisonic decoder for irregular loudspeaker arrays using interaural

cues. In Ambisonics Symposium.

(Cited on page 94)

[291] Tu, W., Hu, R., Wang, H., and Chen, W. (2010). Measurement and

analysis of just noticeable diﬀerence of interaural level diﬀerence cue. 2010

International Conference on Multimedia Technology, pages 1–3.

(Cited on page 148)

[292] Van Wanrooij, M. M. and Van Opstal, A. J. (2004). Contribution of head

shadow and pinna cues to chronic monaural sound localization. Journal of

Neuroscience, 24(17):4163–4171.

(Cited on page 11)

[293] Vorl¨ander, M. (2007). Auralization: Fundamentals of Acoustics, Mod-

elling, Simulation, Algorithms and Acoustic Virtual Reality. RWTHedition.

Springer Berlin Heidelberg.

(Cited on pages 2, 14, 15, 18, 19, 20, 21, 22, 33, 40, 42, 53 and 121)

[294] Vorl¨ander, M. (2008). Virtual Acoustics: Opportunities and limits of

spatial sound reproduction for audiology. Hausdeshoerens-Oldenburg.

(Cited on page 56)

[295] Vorl¨ander, M. (2014). Virtual acoustics. Archives of Acoustics, vol.

39(No 3):307–318.

(Cited on page 40)

[296] Wallach, H. (1938). On sound localization. The Journal of the Acoustical

Society of America, 10(1):83–83.

(Cited on page 10)

BIBLIOGRAPHY 222

[297] Wang, D. and Brown, G. J. (2006). Binaural sound localization. In

Computational Auditory Scene Analysis: Principles, Algorithms, and Ap-

plications, pages 147–185. Wiley.

(Cited on page 146)

[298] Wanner, L., Blat, J., Dasiopoulou, S., Dom´ınguez, M., Llorach, G., Mille,

S., Sukno, F., Kamateri, E., Vrochidis, S., Kompatsiaris, I., Andr´e, E., Lin-

genfelser, F., Mehlmann, G., Stam, A., Stellingwerﬀ, L., Vieru, B., Lamel,

L., Minker, W., Pragst, L., and Ultes, S. (2016). Towards a multimedia

knowledge-based agent with social competence and human interaction ca-

pabilities. In Proceedings of the 1st International Workshop on Multimedia

Analysis and Retrieval for Multimodal Interaction, MARMI ’16, page 21–26,

New York, NY, USA. Association for Computing Machinery.

(Cited on pages 1 and 57)

[299] Ward, D. B. and Abhayapala, T. D. (2001). Reproduction of a plane-

wave sound ﬁeld using an array of loudspeakers. IEEE Transactions on

Speech and Audio Processing, 9(6):697–707.

(Cited on pages 31, 77 and 86)

[300] Wendt, D., Dau, T., and Hjortkjær, J. (2016). Impact of background

noise and sentence complexity on processing demands during sentence com-

prehension. Frontiers in Psychology.

(Cited on page 51)

[301] Wendt, D., Hietkamp, R. K., and Lunner, T. (2017). Impact of noise

and noise reduction on processing eﬀort: A pupillometry study. Ear and

Hearing.

(Cited on pages 50 and 101)

[302] Wendt, D., Koelewijn, T., Ksi¸a˙zek, P., Kramer, S. E., and Lunner,

T. (2018). Toward a more comprehensive understanding of the impact of

masker type and signal-to-noise ratio on the pupillary response while per-

forming a speech-in-noise test. Hearing Research, pages 1–12.

(Cited on pages 50, 101 and 102)

[303] Westermann, A. and Buchholz, J. M. (2017). The eﬀect of nearby

maskers on speech intelligibility in reverberant, multi-talker environments.

The Journal of the Acoustical Society of America, 141(3):2214–2223.

(Cited on pages 42 and 99)

[304] Whitmer, W. M. and Akeroyd, M. A. (2013). The sensitivity of hearing-

impaired adults to acoustic attributes in simulated rooms. Proceedings of

Meetings on Acoustics, 19(1):015109.

(Cited on pages 1, 18 and 50)

[305] Whitmer, W. M., Seeber, B. U., and Akeroyd, M. A. (2012). Apparent

auditory source width insensitivity in older hearing-impaired individuals.

The Journal of the Acoustical Society of America, 132(1):369–379.

(Cited on pages 16, 18 and 40)

[306] Wightman, F. L. and Kistler, D. J. (1992). The dominant role of low-

frequency interaural time diﬀerences in sound localization. The Journal of

the Acoustical Society of America, 91(3):1648–1661.

(Cited on page 10)

BIBLIOGRAPHY 223

[307] Wightman, F. L. and Kistler, D. J. (1997). Monaural sound localization

revisited. The Journal of the Acoustical Society of America, 101(2):1050–

1063.

(Cited on page 11)

[308] Wilcox, R. (2004). Inferences based on a skipped correlation coeﬃcient.

Journal of Applied Statistics, 31(2):131–143.

(Cited on page 116)

[309] Williams, G. (1999). Fourier Acoustics: Sound Radiation and Nearﬁeld

Acoustical Holography. Academic Press.

(Cited on pages 28 and 229)

[310] Wisniewski, M. G., Thompson, E. R., and Iyer, N. (2017). Theta- and

alpha-power enhancements in the electroencephalogram as an auditory de-

layed match-to-sample task becomes impossibly diﬃcult. Psychophysiology,

54(12):1916–1928.

(Cited on page 111)

[311] Wisniewski, M. G., Thompson, E. R., Iyer, N., Estepp, J. R., Goder-

Reiser, M. N., and Sullivan, S. C. (2015). Frontal midline θ power as an

index of listening eﬀort. Neuroreport, 26(2):94—99.

(Cited on page 111)

[312] Wong, G. S. K. (1986). Speed of sound in standard air. The Journal of

the Acoustical Society of America, 79(5):1359–1366.

(Cited on page 15)

[313] W¨ostmann, M., Lim, S.-J., and Obleser, J. (2017). The Human Neu-

ral Alpha Response to Speech is a Proxy of Attentional Control. Cerebral

Cortex, 27(6):3307–3317.

(Cited on page 111)

[314] Xie, B. (2013). Head-related transfer function and virtual auditory dis-

play. J. Ross Publishing.

(Cited on pages 22 and 70)

[315] Yost, W. (2013). Fundamentals of Hearing: An Introduction. Brill.

(Cited on page 8)

[316] Zapata Rodriguez, V., Jeong, C.-H., Hoﬀmann, I., Cho, W.-H., Beldam,

M.-B., and Harte, J. (2019). Acoustic conditions of clinic rooms for sound

ﬁeld audiometry. In Proceedings of 23rd International Congress on Acous-

tics, pages 4654–59. Deutsche Gesellschaft f¨ur Akustik. 23rd International

Congress on Acoustics , ICA 2019 ; Conference date: 09-09-2019 Through

13-09-2019.

(Cited on pages 122 and 139)

[317] Zekveld, A., Kramer, S., and Festen, J. (2011). Cognitive load during

speech perception in noise: The inﬂuence of age, hearing loss, and cognition

on the pupil response. Ear and hearing, 32:498–510.

(Cited on pages 1 and 55)

BIBLIOGRAPHY 224

[318] Zekveld, A. A. and Kramer, S. E. (2014). Cognitive processing load

across a wide range of listening conditions: Insights from pupillometry. Psy-

chophysiology.

(Cited on pages 50 and 112)

[319] Zekveld, A. A., Kramer, S. E., and Festen, J. M. (2010). Pupil response

as an indication of eﬀortful listening: The inﬂuence of sentence intelligibility.

Ear and Hearing.

(Cited on pages 50 and 118)

[320] Zhang, W., Samarasinghe, P., Chen, H., and Abhayapala, T. (2017). Sur-

round by Sound: A Review of Spatial Audio Recording and Reproduction.

Applied Sciences, 7(5):532.

(Cited on pages 19, 20, 21, 27 and 40)

[321] Zobel, B. H., Wagner, A., Sanders, L. D., and Ba¸skent, D. (2019). Spatial

release from informational masking declines with age: Evidence from a de-

tection task in a virtual separation paradigm. The Journal of the Acoustical

Society of America, 146(1):548–566.

(Cited on pages 16 and 40)

[322]

Luboˇs Hl´adek, Ewert, S. D., and Seeber, B. U. (2021). Communication

conditions in virtual acoustic scenes in an underground station.

(Cited on page 42)

[323] S¸aher, K., Rindel, J. H., Nijs, L., and Van Der Voorden, M. (2005).

Impacts of reverberation time, absorption location and background noise

on listening conditions in multi source environment. In Forum Acusticum

Budapest 2005: 4th European Congress on Acustics.

(Cited on page 50)

Appendix A

ITDs Ambisonics

Figure A.1, depicts ITDs for measurements with a listener (HATS manikin)

in the center with Ambisonics (black line), in nine oﬀ-center positions combi-

nations accompanied by a second listener (KEMAR) and alone in those three

oﬀ center positions.

Figure A.1: ITD as a function of source angle in Ambisonics virtualized setup. Top

left HATS displacement = 25 cm, top right HATS displacement = 50 cm, bottom left

HATS displacement = 75 cm, bottom right HATS displacement matching KEMAR

displacement.

225

Appendix B

Delta ILD Ambisonics

Figures B.1, B.2, and B.3, present the diﬀerences in ILD between center and

oﬀ-center listener positions utilizing 24 loudspeakers to render an Ambisonics

with a second listener present inside the ring of loudspeakers. In the ﬁgures,

the number following H indicates the position of the main listener, while the

numbers after K indicate the position of the second listener.

Figure B.1: Diﬀerences in the ILD between centered setup and oﬀ-center setups:

HATS at 25 cm to the right with: KEMAR at 25 cm to the left (top); KEMAR at

50 cm to the left (middle); KEMAR 75 cm to the left (bottom).

226

227

Figure B.2: Diﬀerences in the ILD between centered setup and oﬀ-

center setups: HATS at 50 cm to the right with: KEMAR at 25 cm to

the left (top); KEMAR at 50 cm to the left (middle); KEMAR 75 cm to

the left (bottom).

Figure B.3: Diﬀerences in the ILD between centered setup and oﬀ-

center setups: HATS at 75 cm to the right with: KEMAR at 25 cm to

the left (top); KEMAR at 50 cm to the left (middle); KEMAR 75 cm to

the left (bottom).

Appendix C

Wave Equation and Spherical

Harmonic Representation

Spherical harmonics (SH) represent spatial variations of an orthogonal set

of solutions in the Laplace equation (orthonormal basis) when the solution

is expressed in spherical coordinates, thus giving the spatial representation of

weighted sums in spherical forms that represents a signal (space and frequency

dependent).

C.1 Wave Equation in Spherical Coordinates

Expressing the wave equation in spherical coordinates (r, ϕ, θ) [36] we have:

∂

∂r

∂p

∂r

sin(θ)

∂

∂θ



sin(θ)

∂p

∂θ



sin

(ϕ)

∂

∂ϕ

−

∂

∂t

= 0 , (C.1)

C.2 Separation of the Variables

The diﬀerential equation solution tool called separation of variables can be

used for the Equation C.1, being formulated from the product of three space

dependent variables and a time dependent variable:

p(r, θ, ϕ, t) = R(r)Θ(θ)Φ(ϕ)T (t) . (C.2)

228

Chapter C. Separation of the Variables 229

With the separation of the variables, according to Skudrzyk [278], there are

four homogeneous diﬀerential equations:

dϕ

+ m

= 0 , (C.3a)

sin θ

dθ



sin θ

dΘ

dθ





n(n + 1) −

sin



Θ = 0 , (C.3b)





+ k

R −

n(n + 1)

R = 0 , (C.3c)

+ k

T = 0 . (C.3d)

where m and n integers, the general solutions to the equations are

Φ(ϕ) = Φ

jmϕ

+Φ

− jmϕ

, (C.4a)

Θ(θ) = Θ

(cos(θ)) + Θ

(cos(θ)) , (C.4b)

R(r) = R

(1)

(kr) + R

(2)

(kr) , (C.4c)

T (ω) = T

jωt

− jωt

, (C.4d)

where h

(1)

(x) and h

(2)

(x) are the ﬁrst and second-kind spherical Hankel func-

tions that represent convergent and divergent waves depending on the signal

agreed for the time and P

(x) and Q

(x) are the associated Legendre func-

tions of the ﬁrst and second type.

Due to the singularities in the poles of Legendre’s associated functions at θ = 0

and θ = π the term Θ

is treated as null, and for simpliﬁcation, you can use

the positive m variable or negative, so the term Φ

is also null. According

to Williams [309], for there to be no singularities in the poles of Legendre’s

associated functions, the n index must be an integer. Still, considering causal

systems, the term T

in C.4d is equal to 0 given the convention used.

Chapter C. Spherical Harmonics 230

The associated Legendre functions of the ﬁrst type deﬁned for positive degrees

m are

(x) = (1)

(1 − x

)

(x) . (C.5)

Meanwhile, the functions for negative degrees −m are given by

−m

= (−1)

(n − m)!

(n + m)!

(x) , (C.6)

being the Legendre Polynomial given by

(x) =

− 1)

. (C.7)

C.3 Spherical Harmonics

Equations C.4a and C.4b admit periodic solutions in angular coordinates, and

combined are called spherical harmonics of order n and degree m deﬁned by

(θ, ϕ) =

(2n + 1)

4π

(n − m)!

(n + m)!

(cos(θ)) e

jmϕ

. (C.8)

The negative order SH functions are obtained through the relation

(θ, ϕ) = (−1)

· (Y

−

m(θ, ϕ))

∗

, (C.9)

where

∗

denotes the conjugate complex, and demonstrates that only the phase

changes between the positive and negative degrees of the function. Thus the

magnitude is commonly expressed with the radius and the phase in terms of

color or color scale, as in Figure 2.9.

Appendix D

Reverberation time in Acoustic

Simulation

The reverberation time for the classroom and restaurant are presented in Fig-

ure D.1

(a) (b)

Figure D.1: Reverberation time (a) Classroom (b) Restaurant in octave bands

231

Appendix E

Alpha Coeﬃcients

Figures E.1, E.2, and E.3 presents the absorption coeﬃcients according to the

frequency introduced in the ODEON software to simulate the environments.

Figure E.1: Classroom alpha coeﬃcients (ODEON software).

232

233

Figure E.2: Restaurant alpha coeﬃcients (ODEON software). Figure E.3: Anechoic room alpha coeﬃcients.

Appendix F

Questionnaire

1 | 1

TS_00 Date: ___ / ___ / 2019

Hvor meget anstrengte du dig for at høre sætningerne?

Hvor mange af ordene tror du, at du forstod korrekt?

Hvor ofte måtte du opgive at forstå sætningen?

Ingen

anstrengelse

Lav

anstrengelse

Moderat

anstrengelse

Meget h

anstrengelse

Ingen

Mindre end

halvdelen

Halvdelen

Mere end

halvdelen

Alle

Aldrig

Mindre end

halvdelen af tiden

Halvdelen af

tiden

Mere end

halvdelen af tiden

Altid

234