Objective and perceptive evaluations ofhigh-resolution room acoustic simulations andauralizations

Brian F.G. Katz, David Poirier-QuinotSorbonne Université, CNRS, Institut Jean Le Rond d'Alembert, F-75005 Paris, France.

Barteld N.J. Postma, David TheryLIMSI, CNRS, Université Paris-Saclay, 91405 Orsay, France.

Paul LuizardAudio Communication Group, TU-Berlin

Abstract

Advances in computational power have opened the doors to higher resolution acoustic modelling

for large-scale spaces where acoustics is crucial and spaces are increasingly complicated. As such,

auralizations are becoming more prevalent in architectural acoustics and virtual reality. However,

there have been few studies examining the perceptual quality achievable by room acoustic simulations

and auralizations. This paper presents a summary of several recent studies involving the evaluation,

objectively with regards to acoustic parameters, and perceptively through listening tests, of room

acoustic simulations where subjective equivalency to reality was the driving force. Presented studies

involve the elaboration of a calibration method for simulations, inclusion of dynamic source directivity

characteristics, and the assessment of various simulation methodologies in the context of coupled

volumes. These studies were carried out using existing spaces in order to have a real reference. Room

types included a simpli�ed scale model, a small ornate 570 seat theatre, a 22 200m3 church, and a

84 000m3 cathedral. Results show that state-of-the-art high performance ray/cone tracing simulations

are capable of providing objective and perceptual results, including spatial parameters, comparable to

reference measurements. However, not all algorithms or alternate simulation methodologies provided

equivalent results.

PACS no. 43.55.-n, 43.55.Gx, 43.55.Ka

1. Simulation of complex spaces, the

case of coupled volumes

Coupled volumes are an example of realistic complexspaces as compared to an ideal Sabinian space. In per-forming arts, coupled volume concert halls have beenof increasing interest during the last decades and sev-eral venues have been built with this principle. Thisarchitectural choice provides variable acoustics andinteresting features such as a high sense of sound clar-ity while keeping an important impression of reverber-ation for the audience. This particularity is due to thenon-exponential sound energy decay generated by thedi�erence of reverberation time in the main room andthe acoustic control chamber, meaning that simpli�edsingle room statistical models cannot be applied. The

(c) European Acoustics Association

latter can act as a giant absorber or as a reverbera-tor depending on its own reverberation as comparedto the one in the main audience room. However, thistype of concert hall does not always work as well as de-sired in terms of control chamber e�ciency. Therefore,acoustic simulations in the context of coupled spacesis of primary importance in designing such spaces.Alternatively, the ability of a numerical simulation tomodel such acoustic conditions is an important testcase regarding its viability in the simulation of com-plex room acoustic conditions.

A recent study [1] was carried out comparing sev-eral numerical simulation methods, using physicalmeasurements carried out in a coupled space scalemodel, as a reference. The chosen geometry is a verysimple, schematic coupled system composed of tworooms with di�erent reverberation times acousticallylinked by a single aperture (see Fig. 1a). The reverber-ation times in the two rooms, uncoupled, were de�nedfor each simulation method (see Table 1) using iden-

Copyright © 2018 | EAA – HELINA | ISSN: 2226-5147 All rights reserved

- 2107 -

Table I. Target reverberation times (s) in each indepen-dent room, before coupling, determined from scale modelmeasurements, rescaled to 1:1.

Octave band (Hz) 125 250 500 1000 2000

Main room T30 1.99 1.60 1.53 1.22 0.90Reverberation chamber T30 13.88 10.63 8.48 5.01 3.21

(a)

(b)

0 1 2 3 4−80

−70

−60

−50

−40

−30

−20

−10

0

Time (s)

Ene

rgy

(dB

)

Scale modelRay−tracer 1Ray−tracer 2BEMFDTD

Figure 1. (a) Coupled volume geometry. (b) Time-energydecay curves in the 125Hz 1/3-octave band for impulseresponses from various methods.

tical geometries. Boundary element method (BEM),�nite-di�erence time-domain (FDTD), and two com-mercial geometrical ray-tracing methods were appliedto a common 3D-model representing the actual scalemodel, resized to scale 1:1. Initial con�gurations (un-coupled) were the same for the methods and the re-sulting impulse responses allow for comparison of ac-curacy of acoustical parameters, speci�cally early andlate rates of energy decay.While surface materials in each method were ad-

justed to arrive at the prescribed calibration rever-beration times, once the coupling aperture was in-troduced, the methods signi�cantly varied in theirresults. Examining these results in the 125Hz 1/3-octave band, where wave-based methods are expectedto perform well, the resulting Schroeder energy de-cay curves (see Fig. 1b) show that one of the twocommercial geometrical acoustics methods correctlyreproduced the coupled energy e�ects between thetwo volumes. The other geometrical acoustics soft-ware, as well as the BEM and FDTD methods failedto correctly model the late reverberation in this sim-

pli�ed complex geometry. The use of statistical late-reverberation in some geometrical acoustics solvershinders their ability to correctly model complex ge-ometries. While wave-based methods have been shownto be very reliable for short impulse response di�rac-tion studies [2, 3], these results indicate that thereare issues to be resolved regarding longer propaga-tion times where numerical artifacts can accumulate.Further analysis of the results and discussion can befound in [1].

2. Numerical simulation calibration

As a means of establishing con�dence in the resultsof numerical room acoustic simulations, speci�callywhen employed for auralizations, we are interested inevaluating simulations based on existing buildings, ascompared to auralizations based on measured roomimpulse responses (RIR). While also of a basic funda-mental interest with regards to the quality of simula-tions, there are also direct applications to such com-parisons such as in virtual reality constructions of ex-isting spaces, research on historical structures for ex-amining the e�ects of renovations, or conversely inthe architectural acoustics conceptualization and de-sign of renovations to existing spaces.

According to [4], a simulation model is well cali-brated when the di�erence between model and mea-surement is less than the Just Noticeable Di�erence(JND). If uncertainties are smaller than these val-ues, the simulation can be considered as perceptu-ally equivalent. An alternate means of validating themodel uses subjective listening tests where simulatedauralizations are compared to real life recordings [5].

A number of recent studies have examined the cre-ation of calibrated acoustic of historical sites. A studyof the Fogg Art Museum [6] in which Wallace C.Sabine did his �rst tests on reverberation produceda calibrated historic model validated according toacoustic parameters. [7] studied the change in acousticconditions of the �Misteri d'Elx" which was performedin the Basilica �Santa Maria de Elche". Models of theBasilica were simulated and calibrated based on in-situ measurements of material absorption coe�cients.[8] studied historic mosques and byzantine churchesin Istanbul. A recording from a live performance wascompared to a simulated impulse response convolvedwith anechoic recordings.

2.1. Objective calibration

To minimize the di�erences between measured andsimulated results, a method has been proposed in[9, 10] to objectively calibrate a geometrical acousticsmodel to measured data of a real space, resulting ina methodical calibration procedure. This calibrationprocedure was validated by means of objective pa-rameter comparisons for omni-directional source and

Euronoise 2018 - Conference Proceedings

- 2108 -

Table II. Approximate details and dimensions of the Ge-ometrical models of the Saint-Germain-des-Prés church(SGdP), Notre-Dame cathedral (ND), and Théâtre del'Athénée.Room Polygons Volume (m3) Floor plan area (m2)

Saint-Germain-des-Prés 2,200 22,200 1,800Notre-Dame 14,700 84,000 4,800Théâtre de l'Athénée 1,300 2,500 300

receivers. Three architectural sites with critical acous-tics in Paris, France, were tested using this calibra-tion procedure: the abbey of Saint-Germain-des-Prés,Notre-Dame cathedral, and the Théâtre de l'Athénée.

Impulse response measurements were carried outin the three sites according to the measurement plansshown in Figures 2(a�c). The Exponential Swept Sinemethod [11] was employed. The sweep frequency wentfrom 20 Hz to 20 kHz, with a duration of 20 s in Notre-Dame and 10 s for the other two rooms. The sweep wasplayed and recorded using the DAW software Reaperand sound card (RME Fireface 800). Deconvolutionsteps were carried out in MatLab to obtain the RIR.The audio output was sent to an ampli�er (SAMSOMServo 120a) and two miniature dodecahedral soundsources (model 3D-032, Dr-Three). Omnidirectionalmicrophones (model 4006, DPA) were employed forcalibration positions. An arti�cial head (Neuman KU80 equipped with model 4060, DPA) was used for thelistening test positions as indicated.

CATT-Acoustic (v.9.0.c:3, TUCT v1.1a:4) was em-ployed to create the geomatrical acoustic models andperform simulations. The geometries of the Notre-Dame cathedral and Saint-Germain-des-Prés churchwere determined from available 3D laser scan pointclouds as well as architectural plans and sections.The geometry of the Théâtre de l'Athénée was de-termined from architectural plans and sections. Ta-ble II presents a summary of the geometrical modeldetails. Surface materials were determined from vi-sual inspection. Initial absorption coe�cients wereadopted from publicly available databases [12, 13, 14].Scattering coe�cients were generally modeled usingthe CATT-Acoustic option estimate which provides asimple estimation of this frequency dependent coe�-cient based on a given characteristic depth representa-tive of the surface's roughness. The choice of calcula-tion algorithm depends on the geometry of the space.As the Notre-Dame cathedral and Saint-Germain-des-Prés church have fairly even absorption and rever-beration times which are not strongly dependent onscattering, simulations were run with Algorithm 1:Short calculation, basic auralization and transition or-der 1 with 250,000 and 150,000 rays, respectively. Incontrast, absorption distribution in the Théâtre del'Athénée is not uniform. As such, simulations wereperformed using Algorithm 2: Longer calculation, de-tailed auralization with 100,000 rays.

Calibration of the geometrical acoustic models fol-lowed a 7-step calibration procedure, proposed in [9],is summarized here:

1. RIR measurements are carried out in the studiedvenue. The results of these measurements are usedas a reference for the calibration.

2. The geometrical model is created and the geometryremains unchanged during calibration.

3. Preliminary (but realistic) acoustical properties areassigned to all surfaces, resulting in a geometricalacoustic model.

4. Since stochastic implementations of Lambert scat-tering in geometrical acoustic software leads to run-to-run variations, the geometrical acoustic model'srepeatability is quanti�ed. These variations arethen taken into account when simulations and mea-surements are compared.

5. The sensitivity of the geometrical acoustic modelto variations of scattering coe�cients is quanti�ed.For this purpose, simulations are run of the initialgeometrical acoustic model followed by simulationswith all scattering coe�cients set to 0%, then to99%, with absorption coe�cients unchanged.

6. Acoustical surface properties (absorption and scat-tering coe�cients) are adjusted, taking into ac-count the determined sensitivities, to arrive atglobal mean di�erences between measurement andsimulated results for reverberation and clarity pa-rameters within a tolerance threshold (ε).

7. Acoustic properties of local key surfaces are ad-justed to minimize the standard deviation (SD) ofpairwise di�erences in reverberance and clarity pa-rameters.

For simplicity and uniformity in this study, the ISO3382 standard's [15] JNDs (JNDEDT = 5%, range= 1.0− 3.0 s; JNDC80 = 1 dB, range = ±5 dB) wereselected as calibration target thresholds (ε) for the dif-ferent parameters. It is noted that these JND valuesvary as a function of room usage, but they are cho-sen as a base model tolerance reference value for thepurpose of this study. In a speci�c instance, a moresuitable threshold should be used which is speci�callyappropriate to the room's function.

During step 6 of the calibration procedure, rever-beration parameters T20 and EDT were calibratedby adjusting the absorption coe�cients before adjust-ing scattering coe�cients to calibrate clarity param-eters C50 and C80. This step was performed withthe baseline requirement to keep the material prop-erties within physically viable values. To provide anoverview of the di�erence between measurement andcalibrated geometrical acoustic models, Figures 2d�f compare the mean measured EDT and C50 of allsource-receiver combinations for the omni-directionalmicrophones to those of the simulations. Simulatedreverberation parameter EDT is within 1ε of the mea-sured values across all frequency bands for the three

Euronoise 2018 - Conference Proceedings

- 2109 -

(a) (b) (c)

(d) (e) (f)

Figure 2. Measurement plans (not to same scale) of (a) Saint-Germain-des-Prés church, (b) Notre-Dame cathedral, and(c) Théâtre de l'Athénée. S and R represent source and receiver positions (S# and R# were employed in the listeningtest). Comparison between simulated and measured mean (±1ε) EDT and C50 for the (d) Saint-Germain-des-Prés church,(e) Notre-Dame cathedral, and (f) Théâtre de l'Athénée.

rooms. The Saint-Germain-des-Prés model overesti-mated the C50 by slightly more than 1ε in the 2000 Hzoctave band. The Notre-Dame model overestimatedthe C50 by slightly more than 1ε in the 500 and2000�4000 Hz octave bands. The Théâtre de l'Athénéemodel estimated the C50 within 1ε of the measuredvalue across all octave bands. T20 results were similarto EDT results and C80 results were comparable toC50 results. The slight overestimation of clarity pa-rameters could not be corrected while both maintain-ing reverberation parameter calibration and keepingscattering properties within physically viable values.

2.2. Perceptual listening test evaluation

While the results of Sec. 2.1 show the ability of nu-merical geometrical acoustic simulations to provide

accurate acoustic parameter reproduction, this doesnot quanlify as a validation of perceptual equivalence.To carry out reliable auralization comparisons, be-sides acoustic parameter comparisons, one needs toascertain that other elements of the simulation arecon�gured comparable to the measurement as well.We give speci�c attention here to binaural auraliza-tions as they provide the most natural reconstruc-tion of the listening experience. The use of a di�er-ent source or receiver directivity or binaural HRTFcan lead to coloration and other signi�cant perceptualdi�erences. On the other hand, preparation steps arerequired for the measured Binaural Room Impulse Re-sponse (BRIR) as well. Compensation must be madefor the frequency response characteristics of the mea-surement system and for di�erences in signal-to-noise

Euronoise 2018 - Conference Proceedings

- 2110 -

Table III. Control factors between measurement andsimulation in previous auralization comparison studies.A `X' indicates that the factor was controlled (i.e.equal) between measurement/recording and simulation, a`×'indicates a di�erence. † this control factor was equalfor the noisy conditions, however di�ered for the noise freeconditions.Factor Lokki[16, 17] Choi[18] Yang[19]

Source directivity X X XMeas. system's freq. response × X ×SNR X × †HRTFs X × ×

ratio (SNR) between measurement and simulation asthe simulated BRIRs is free of background-noise.A select number of comparisons of binaural aural-

izations have previously been carried. [16, 17] com-pared real-head binaural recordings to simulated au-ralizations of a lecture room using subjective listen-ing tests. [18] attempted to validate computer mod-els of two concert halls by comparing recordings tosimulated auralizations. [19] attempted to validatecomputed auralizations for use in speech-intelligibilitystudies. Table III provides an overview of which fac-tors were controlled between measurement and sim-ulation in these binaural evaluation studies. Resultsof these studies have highlighted potential problemsin creating numerical models to obtain perceptuallyvalid auralizations. First, several control factors dif-fered between measurement and simulation. Second,objective and subjective di�erences were identi�ed be-tween the measurement and simulation, clarity pa-rameters seemed especially problematic.For the current study, the resulting calibrated ge-

ometrical acoustic models from Sec. 2.1 were eval-uated via paired comparison auralizations betweenmeasured and simulated BRIRs. Three site appropri-ate stimuli, each with a duration of∼13 s were used forthe auralizations. RMS of the measured and simulatedconvolutions was used for level normalization. For theNotre-Dame cathedral and Saint-Germain-des-Préschurch: a female soprano singing `Abendemp�ndung',by W.A. Mozart, for details of the anechoic record-ing system see [20]. For the Théâtre de l'Athénée: aFrench speaking male reciting a translated extract of`Hamlet', by W. Shakespeare, and a Italian speak-ing male reciting an extract of `Non recidere, for-bice, quel volto' by E. Montale. These anechoic stimuliwere recorded in the anechoic room (IRCAM, Paris)using an omni-directional microphone (model 4006,DPA) at 4m distance. Additional details of the studycan be found in [10], including comparisons of omni-directional and spatial acoustic parameters.26 participants took part in the study. They were

not informed of the nature of the recordings (i.e. sim-ulated or measured). They were asked to rate the sim-ilarity of samples according to 8 perceptual attributes:

• Reverberance (reverb)

Figure 3. Subject results on similarity of measured andsimulated binaural auralizations. Center dashed line rep-resents a neutral response. Dashed vertical bars representindividual attribute repeatability mean values.

• Clarity• Distance (dist.)• Tonal balance (ton. bal.)• Coloration (col.)• Plausibility (plaus.) - Given the assumption that

the binaural recordings were made in a church(singing voice) or in a theatre (spoken voice): Doesthe recording sound reasonable to you?

• Apparent Source Width (ASW)• Listener EnVelopment (LEV)

Participants responded using a continuous graphicslider scale (100 pt), with the end parts labeled `A ismuch more . . . ' and `B is much more . . . ' correspond-ing to end values of �50 and +50 respectively, with acenter 0 response indicating no perceived di�erence.Presentation order and AB correspondence to simula-tion and measurement were randomized. Participantswere able to listen to the compared pairs as manytimes as desired. Auralizations were presented viaheadphones (Sennheiser model HD 600) at an RMSlevel of 75 dBA.Combined results are compared (see `total' box-

plot in Fig. 2.2). All perceptual attribute results werewithin the repeatability tolerance ranges. No percep-tual di�erences were observed for the attributes rever-berance, tonal balance, coloration, plausibility, ASW,and LEV. However, the simulated auralizations of theSaint-Germain-des-Prés church were judged �clearer�and �closer� than the measured auralizations, whileThéâtre de l'Athénée and Notre-Dame judgmentswere within attribute repetition tolerances.These combined results show that with a current

commercially available geometrical acoustics soft-ware (speci�cally CATT-Acoustic (v.9.0.c:3, TUCT

Euronoise 2018 - Conference Proceedings

- 2111 -

v1.1a:4) in these studies), employing a methodical cal-ibration approach based on few parameters evaluatedfor omni-directional source-receiver pairs, one is capa-ble of producing perceptually equivalent spatial bin-aural auralizations when compared to measured bin-aural responses.

3. Perception of dynamic source di-

rectivity

In numerical simulations, it is often possible to pre-scribe the directivity of an acoustic source in order tobetter represent the way in which it excites the room.Such directivities are typically static, being de�nedaccording to source excitation as a function of fre-quency. However, real voice directivity varies due toboth phoneme dependent radiation patterns [21] anddynamic orientation of the talker/speaker.Previous studies [22, 23] have proposed to achieve

the inclusion of dynamic directivity through the useof multi-channel source directivity auralization1. Thismethod employs anechoic multi-channel recordings.The source's radiation sphere is divided into segmentsrepresenting each microphone position. The RIR isthen calculated for each segment and convolved withthe corresponding microphone channel of the ane-choic recording. Convolutions of each channel are thendown-mixed to create a multi-channel source directiv-ity auralization.As segmented directivity approaches lead to dis-

crete and abrupt level changes when source orien-tation is altered, [24, 25] proposed a multi-channelsource decomposition using an overlapping beamform-ing approach. To accomplish this, the radiation sphereis decomposed in 12 equally distributed beam pat-terns. The beams were designed to have minimaloverlap while also resulting in an equal gain sumfor all directions in order to reproduce an omni-directional pattern. The following control points wereemployed: 0◦ no attenuation, 42◦ (dodecahedron ver-tices) 3 beams sum to 0 dB, and 90◦ maximum atten-uation.The 2D beam was generated using a spline inter-

polation (5◦ steps, rotated around its symmetry axisto create the 3D beam. The 12 beam instances wereorientated towards each face of a dodecahedron. Theresult of the equal-weighted summation accurately re-produced an omni-directional sphere (±0.3 dB). Ad-ditional details can be found in [24, 25].The 12-beam source was positioned at center stage

of the previously calibrated geometrical acousticsmodel of the Théâtre de l'Athénée. Three receiverswere positioned in the audience on the main �oor.

1 The original paper termed this application �multi-channel au-ralization�. In order to avoid confusion with distributed sourcemulti-channel auralizations, we use the term multi-channel

source directivity auralization.

Simulations were carried out using CATT-Acoustic(v.9.1, TUCT v2.0, 200,000 rays, Algo. 3 ), produc-ing 2nd order Ambisonic RIRs.Voice directivity variations were controlled via gain

modulations of the 12-beam components. Frequencydependent gain weighting factors were determinedbased on measured voice directivity data [26]. Thisdata was converted to octave band values and thenencoded using 3rd order spherical harmonic decom-position to facilitate interpolation.To provide an `anechoic' sound �le with corre-

sponding source orientation information, a perfor-mance with two actors (�Ubu Roi� by Alfred Jarry)was recorded with head-worn mics and simultaneously�lmed using a Kinect 2 RGB/Depth sensor in a smalldry theater (see [27] for more details). The close-micrecordings were employed in analogy with anechoicrecordings. A 17 s extract of this performance wasselected which highlighted variations in actors' headorientation and movement during the scene. Actors'head orientation were tracked.A real-time Max/MSP patch was designed which

modulated the gains of the 12-beam RIR contri-butions to create the desired directivity. First, the`anechoic' recording was convolved with the 12-beamsource RIRs. These convolved channels were then �l-tered in octave bands 125�4000 Hz. Gain weightingfactors were calculated by �rst applying any source ro-tation in the spherical harmonic domain. The octave-band rotated directivity data was then decoded tothe 12-beam directions, resulting in a set of weightingfactors which were applied to the convolved �lteredchannels which were subsequently summed. For anomni-directional source, all weightings factors wereset to unity. For the static voice directivity auraliza-tion, the source orientation was �xed towards the au-dience. For dynamic voice directivity, source rotationused the actors' head orientation. These convolutionswere decoded from 2nd order Ambisonics to binau-ral, employing Spat 's2 virtual speaker array approach(18 speakers uniformly distributed de�ned) and theprovided KEMAR HRTF, to produce the �nal bin-aural auralizations. The di�erent auralizations werethen recorded and RMS level normalized.The resulting auralizations were compared by

means of a listening test. The test had 3 variants cor-responding to the source directivity-types. Binauralauralizations were compared for 3 receiver positions.Every trial was repeated 3 times, resulting in (3×3)9 trials. The 21 participants compared and rated the3 auralizations in terms of Plausibility (Plaus.), Dis-tance (Dist.), ASW, and LEV 3 on a discrete scaleranging from 1 (`least . . . ') to 7 (`most . . . '). Addi-tionally, before the test commenced, participants wereshown the silent RGB video in order to provide them

2 http://forumnet.ircam.fr/product/spat-en/

3 For descriptions provided to the participants see [10]

Euronoise 2018 - Conference Proceedings

- 2112 -

Figure 4. Summary of (a) Total results and individualpositions (b) front, (c) middle, and (d) rear parterre oncenterline. (white) omni-directional, (cyan) static voice,and (red) dynamic voice directivity.

with a contextual reference for the attribute Plausi-bility. A video was selected over an image in order toput the subjects in mind of a real play, with actordynamics included. Auralizations were presented viaheadphones (Sennheiser model HD 650) at an RMSlevel of 80 dBA.A summary of the results are presented in Fig. 3.

For all 3 positions, the dynamic voice sources wereconsidered signi�cantly closer, wider, and the aural-izations more enveloping than the omni-directionalauralizations. Additionally, for the middle and rearparterre positions the dynamic voice auralizationswere perceived more plausible than the omni-directional auralizations. The poor plausibility of thefront parterre listening position was attributed to thefact that the, while in the video the actors were sepa-rate from each other, due to technical constraints bothvoices were auralized from the same stage position.Subsequent studies have remedied this limitation, atthe cost of full real-time implementation.

4. Conclusions

A series of recent studies concerning the accuracy,both via objective parameters and perceptive test-ing, have been presented. In comparing geometri-cal acoustics ray-tracing and wave-based approaches,both BEM and FDTD, in the context of complex en-ergy exchanges between coupled volumes, resulting inmulti-slope energy decay rates has shown that suchwave-based methods, and some ray-trace methods,are not currently capable of modeling such acous-tic phenomenon. As such, their application in roomacoustics remains questionable for any non-ideal sin-gle geometry where late elements of the response needto be calculated. For methods which perform su�-ciently well, it has been shown that following a me-

thodical calibration procedure, �rst minimizing meanerror and then �ne tuning to minimize standard de-viation of errors for reverberation and clarity param-eters, results numerical simulations using geometricalacoustics modeling are capable of recreating perceptu-ally equivalent spatialized auralizations as comparedto measured data. Further enhancing of the numer-ical models through the inclusion of dynamic sourcedirectivity has been shown to be both possible, andbene�cial to the plausibility of resulting auralizations.While these studies have focused on audio-only ren-

derings, it is becoming more and more common tocombine audio auralizations and real-time simulationswith visual graphics. In the context of the above men-tioned studies, annex works have created virtual sce-narios in both Notre-Dame cathedral [28, 29] and theThéâtre de l'Athénée [30, 27]. These works have ledto subsequent studies concerning the impact of visualrendering on acoustic perception, in the context ofroom acoustics. Such studies have examined the ef-fect of acoustic attribute ratings with the addition ofvisual information as well as the impact of incoheren-cies in audio and visual distance [31, 32]. Finally, withthe increasing prevalence of immersive visual systems,these studies are being extended to examine the im-pact of the visual rendering quality (in resolution andimmersion) on the same room acoustic attribute rat-ings [33].

Acknowledgement

Portions of this work were funded by the FrenchFUI project BiLi (�Binaural Listening�, www.bili-project.org, FUI-AAP14). Portions of this work werefunded by the ECHO project: �The emergence ofmodern orality and aurality. Phonic movementsin the scenic image� (ANR-13-CULT-0004, echo-projet.limsi.fr).

References

[1] P. Luizard, M. Otani, J. Botts, L. Savioja, andB. F. Katz, �Comparison of sound �eld measure-ments and predictions in coupled volumes betweennumerical methods and scale model measurements,�in Proc. of Meetings on Acoustics, vol. 19, (Montreal),pp. 015114:1�9, June 2013, doi:10.1121/1.4799138.

[2] B. Katz, �Boundary element method calculationof individual head-related transfer function. II.Impedance e�ects and comparisons to real measure-ments,� J. Acoust. Soc. Am., vol. 110, no. 5, pp. 2449�2455, 2001, doi:10.1121/1.1412441.

[3] R. Gre� and B. Katz, �Round robin comparison ofHRTF simulation results : preliminary results.,� inAudio Eng Soc Conv 123, (New York, USA), pp. 1�5, 2007, (url).

[4] M. Vorländer, �Prediction tools in acoustics - can wetrust the PC?,� in BNAM, 2010.

[5] J. Rindel and C. Christensen, �Room acoustic sim-ulation and auralization: How close can we get tothe real room?,� in Proceeding of the Western Paci�cAcoustics Conference, 2003.

Euronoise 2018 - Conference Proceedings

- 2113 -

[6] B. Katz and E. A. Wetherill, �Fogg art museum lec-ture room, a calibrated recreation of the birthplaceof room acoustics,� in Forum Acusticum, (Budapest),pp. 2191�2196, 2005.

[7] D. P. Garcia, M. Rychtáriková, C. Glorieux, andB. F. Katz, �Interactive auralization of self-generatedoral sounds in virtual acoustic environments for re-search in human echolocation,� in Forum Acusticum,(Krakow), pp. 1�6, Sept. 2014, (url).

[8] C. Weitze, C. Christensen, and J. Rindel, �Computersimulation of the acoustics of mosques and byzantinechurches,� in Proceedings of the ICA, 2001.

[9] B. N. Postma and B. F. Katz, �Creation and calibra-tion method of virtual acoustic models for historicauralizations,� Virtual Reality, vol. 19, no. SI: Spa-tial Sound, pp. 161�180, 2015, doi:10.1007/s10055-015-0275-3.

[10] B. N. Postma and B. F. Katz, �Perceptive and ob-jective evaluation of calibrated room acoustic simu-lation auralizations,� J. Acoust. Soc. Am., vol. 140,pp. 4326�4337, Dec. 2016, doi:10.1121/1.4971422.

[11] A. Farina, �Simultaneous measurement of impulse re-sponse and distortion with a swept-sine technique,�Aud Eng Soc Conv 108, 2000.

[12] L. Beranek, Acoustics. Acoustical Society of America,1986.

[13] L. Beranek, Concert and Opera Halls: How theysound. Acoustical Society of America, 1996.

[14] M. Vorländer, Auralizations Fundamentals of Acous-tics, Modeling, Simulation, Algorithms and AcousticVirtual Reality. Springer-Verlag, 2008.

[15] ISO 3382-1, �Acoustics � Measurement of roomacoustic parameters � Part 1: Performance spaces.�Int. Organization for Standardization, 2009.

[16] T. Lokki and H. Järveläinen, �Subjective evaluationof auralization of physics-based room acoustics mod-eling,� in Intl. Conf. on Auditory Display, pp. 1�6,2001.

[17] T. Lokki, L. Savioja, R. Vaananen, J. Huopaniemi,and T. Takala, �Creating interactive virtual auditoryenvironments,� IEEE Computer Graphics And Appli-cations, vol. 22, no. 4, pp. 49�57, 2002.

[18] Y.-J. Choi and F. Fricke, �A comparison of subjectiveassessments of recorded music and computer simu-lated auralizations in two auditoria,� Acta Acusticaunited with Acustica, vol. 92, pp. 604�611, 2006.

[19] W. Yang and M. Hodgson, �Validation of the aural-ization technique: Comparative speech-intelligibilitytests in real and virtual classrooms,� Acta Acusticaunited with Acustica, vol. 93, no. 6, pp. 991�999, 2007.

[20] T. Lokki, V. Pulkki, and P. T. Calamia, �Measure-ment and modeling of di�raction from an edge of athin panel,� Applied Acoustics, vol. 69, no. 9, pp. 824�832, 2008.

[21] B. F. Katz, F. Prezat, and C. d'Alessandro, �Humanvoice phoneme directivity pattern measurements,� inFourth Joint Meeting: ASA and ASJ, 2006.

[22] J. Rindel, F. Otondo, and C. Christensen, �Soundsource representation for auralization,� in Int Sympon Rm Acoust, (Hyogo, Japan), pp. 1�8, 2004.

[23] F. Otondo and J. Rindel, �A new method for the radi-ation representation of musical instruments in aural-izations,� Acta Acustica, vol. 91, no. 5, pp. 902�906,2005.

[24] B. N. Postma and B. F. Katz, �Dynamic voice direc-tivity in room acoustic auralizations,� in German An-nual Conf. on Acoustics (DAGA), pp. 352�355, Mar2016, (url).

[25] B. N. Postma, H. Demontis, and B. F. Katz, �Subjec-tive evaluation of dynamic voice directivity for aural-izations,� Acta Acust united Ac, vol. 103, pp. 181�184,2017, doi:10.3813/AAA.919045.

[26] W. T. Chu and A. C. C. Warnock, �Detailed directiv-ity of sound �elds around human talkers,� Tech. Rep.IRC-RR-104, NRC Institute for Research in Con-struction, Dec. 2002.

[27] D. Poirier-Quinot, B. N. Postma, and B. F. G. Katz,�Augmented auralization : Complimenting auraliza-tions with immersive virtual reality technologies,� inIntl. Sym on Music and Room Acoustics (ISMRA),(La Plata), pp. 14:1�10, Sept. 2016, (url).

[28] B. N. Postma, D. Poirier-Quinot, J. Meyer, and B. F.Katz, �Virtual reality performance auralization in acalibrated model of Notre-Dame Cathedral,� in Eu-roregio, (Porto), pp. 6:1�10, June 2016, (url).

[29] B. F. Katz, B. N. Postma, D. Poirier-Quinot, andJ. Meyer, �Experience with a virtual reality auraliza-tion of Notre-Dame Cathedral,� vol. 141, (Boston),p. 3454, ASA, 2017.

[30] B. F. G. Katz and B. N. Postma, �L'auralisation aug-mentée de la salle Louis Jouvet du Théâtretre del'Athénée. L'acoustique historique des lieux de spec-tacle,� in La scène parle. Voix, acoustiques et aural-ités en France dans la seconde moitié du XXe siècle,(Paris), Dec. 2016.

[31] B. N. Postma and B. F. Katz, �In�uence of visualrendering on the acoustic judgments of a theater au-ralization,� in Proc. Mtgs. Acoust., vol. 30, (Boston),pp. 015008:1�13, ASA, 2017, doi:10.1121/2.0000575.

[32] B. N. Postma and B. F. Katz, �The in�uence of visualdistance on the room-acoustic experience of auraliza-tions,� J. Acoust. Soc. Am., vol. 142, no. 5, pp. 3035�3046, 2017, doi:10.1121/1.5009554.

[33] D. Thery, D. Poirier-Quinot, B. N. Postma, andB. F. Katz, �Virtual reality and augmented reality,�in Lecture Notes in Computer Science (J. Barbic,M. D'Cruz, M. Latoschik, M. Slater, and P. Bourdot,eds.), no. 10700 (EuroVR 2017) in LNCS, ch. Impactof the Visual Rendering System on Subjective Au-ralization Assessment in VR, pp. 105�118, Springer,2017.

Euronoise 2018 - Conference Proceedings

- 2114 -