effect of acoustic environment on the sensitivity of speech transmission index to source directivity

This article was downloaded by: [Florida Atlantic University]On: 16 November 2014, At: 13:18Publisher: Taylor & FrancisInforma Ltd Registered in England and Wales Registered Number: 1072954 Registered office: MortimerHouse, 37-41 Mortimer Street, London W1T 3JH, UK

Architectural Science ReviewPublication details, including instructions for authors and subscription information:http://www.tandfonline.com/loi/tasr20

Effect of Acoustic Environment on the Sensitivity ofSpeech Transmission Index to Source DirectivityKen Stewart a & Densil Cabrera aa Faculty of Architecture, Design and Planning , University of Sydney , NSW, 2006,AustraliaPublished online: 09 Jun 2011.

To cite this article: Ken Stewart & Densil Cabrera (2009) Effect of Acoustic Environment on the Sensitivity of SpeechTransmission Index to Source Directivity, Architectural Science Review, 52:1, 71-78

To link to this article: http://dx.doi.org/10.3763/asre.2009.0006

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) containedin the publications on our platform. However, Taylor & Francis, our agents, and our licensors make norepresentations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose ofthe Content. Any opinions and views expressed in this publication are the opinions and views of the authors,and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be reliedupon and should be independently verified with primary sources of information. Taylor and Francis shallnot be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and otherliabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to orarising out of the use of the Content.

This article may be used for research, teaching, and private study purposes. Any substantial or systematicreproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in anyform to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://www.tandfonline.com/page/terms-and-conditions

http://www.tandfonline.com/loi/tasr20

http://dx.doi.org/10.3763/asre.2009.0006

http://www.tandfonline.com/page/terms-and-conditions

http://www.tandfonline.com/page/terms-and-conditions

©2009EarthscanLtd.ISSN:0003-0628(print),1758-9622(online)doi:10.3763/asre.2009.0006www.earthscan.co.uk/journals/asre

ArchitecturalScienceReviewVolume52.1,pp71-78

EffectofAcousticEnvironmentontheSensitivityofSpeechTransmissionIndexto

SourceDirectivity

KenStewartandDensilCabrera†

FacultyofArchitecture,DesignandPlanning,UniversityofSydney,NSW2006,Australia†Correspondingauthor:Tel:(+612)93515267;Fax:(+612)93513031;Skype:densilcabrera;

Email:[email protected]

Received13January2009;accepted20January2009

Abstract: Source directivity has the potential to affect speech transmission index (STI) measurements. When the sourceis a model of a human talker, there is a question of how accurate that model needs to be. However, in many acousticenvironments source directivity has little effect on STI because other factors dominate.One instance is soundfields in whichthe direct sound is dominant (e.g., in the quiet outdoors, in the nearfield, or in an anechoic room): directivity will have noeffect on STI so long as the direct sound is strong enough. Another instance is soundfields in which the reverberantsoundfieldisdominant(e.g.,inthefarfieldinaroomwithmoderateormorereverberation).Thispaperexaminestheoreticalsituations where source directivity has a substantial influence on STI because of the balance between direct (or early) andreverberant soundfields, aswell as the role thatbackgroundnoise canhave in increasing the importanceof sourcedirectivity.

Keywords:Architecturalacoustics,Loudspeakers,Speechdirectivity,Speechintelligibility

IntroductionSpeechtransmission index(STI) isanobjective indicator

of speech intelligibility derived from the measurement ofthe modulation transfer function of a system (InternationalElectrotechnical Commission, 2003). It may be used inassessing room acoustics, sound reinforcement systems,telecomunicationssystems,andothersystemsinwhichspeechisconveyedtopeople.Forameasurementoftenaloudspeakerwouldbeusedas the initial soundsource for the test signal– and this loudspeaker should have similar directionalcharacteristics to those of a human talker. However, inpractice loudspeakers that closely match human speechdirectivitymaynotbeavailable,sothequestionarisesofhowsensitive a measurement would be to deviations from idealdirectivity.InthispaperweconsiderthepotentialinfluencethatsoundsourcedirectivityhasonSTIintermsofsource-receiverdistance,roomvolume,roomreverberationtime,andbackgroundnoiselevel.

Previous studies by Bozzoli, Bilzi and Farina, (2005a),Bozzoli,ViktrorovitchandFarina, (2005b)andStewartandCabrera,(2007)haveexploredtheissueofdirectivityofSTIsound sources through physical measurement. By testing avariety of loudspeaker enclosures approximating the humanform in various room acoustical contexts, Bozzoli et al.,

(2005a) found that directivity appeared to have little effectonSTImeasurementsinnormalrooms(classrooms),butdidhaveasubstantialeffectinacarcabin.Bozzoliet al.,(2005b)foundthatthemeasuredaveragedirectivityoftentalkerswasagoodmatchforthedirectivityofaBruelandKjaer4128Cheadandtorsosimulator.However,oneissuewiththisisthatahead and torso simulator is very expensive and somewhatfragile, and so is not necessarily a practical solution to STImeasurement. Stewart and Cabrera, (2007) examined theeffect of mouth size on the directivity of a head and torsosimulator, finding a substantial effect at some frequencies.WhentheeffectofmouthsizeonSTImeasurementsinrealrooms(alecturetheatreandameetingroom)wastested,therewaslittleeffect,withonlyamodestvariationinSTIinoneofthepositionstested.

Thesemeasurement-basedstudiesarelimitedbytheeffortinvolvedinmakingphysicalmeasurements,andsoonlyofferaglimpseoftherangeofpossiblescenarios.Hence,thepresentpapertakesatheoreticalappoachtothequestion.

ModelofSTIAccordingtoHoutgast,Steeneken,andPlomp,(1980),the

modulationtransferfunctionofaroomacousticalsystem,m(F),

Dow

nloa

ded

by [

Flor

ida

Atla

ntic

Uni

vers

ity]

at 1

3:18

16

Nov

embe

r 20

14

ArchitecturalScienceReview Volume52,Number1,200972

canbe estimated from equation1, based onprinciples fromstatisticalroomacoustics.Themodulationtransferfunctionisformedbyasetofmoduationreductioncoefficients(i.e.,valuesthatenumeratetheextenttowhichthemodulationdepthofamodulatedsignalisreducedbytheacousticsystem)forarangeof modulation frequencies (F), which for STI measurementsspan 0.63 Hz to 12.5 Hz in 1/3-octave intervals. In STImeasurements,thismodulationtransferfunctioniscalculatedforeachofsevencarrierfrequencyoctavebands,from125Hzto8kHz.

Proceedings of ACOUSTICS 2008 24-26 November 2008, Geelong, Australia

Acoustics 2008 1

Effect of acoustic environment on the sensitivity of speech transmission index to source directivity

Ken Stewart and Densil Cabrera

Faculty of Architecture, Design and Planning, University of Sydney, NSW 2006, Australia

ABSTRACT

Source directivity has the potential to affect speech transmission index (STI) measurements. When the source is a model of a human

talker, there is a question of how accurate that model needs to be. However, in many acoustic environments source directivity has

little effect on STI because other factors dominate. One instance is soundfields in which the direct sound is dominant (e.g., in thequiet outdoors, in the nearfield, or in an anechoic room): directivity will have no effect on STI so long as the direct sound is strong

enough. Another instance is soundfields in which the reverberant soundfield is dominant (e.g., in the far field in a room with moder-

ate or more reverberation). This paper examines theoretical situations where source directivity has a substantial influence on STI

because of the balance between direct (or early) and reverberant soundfields, as well as the role that background noise can have inincreasing the importance of source directivity.

INTRODUCTION

Speech transmission index (STI) is an objective indicator of speech intelligibility derived from the measurement of the modulation

transfer function of a system (International Electrotechnical Commission 2003). It may be used in assessing room acoustics, sound

reinforcement systems, telecomunications systems, and other systems in which speech is conveyed to people. For a measurement

often a loudspeaker would be used as the initial sound source for the test signal – and this loudspeaker should have similar directionalcharacteristics to those of a human talker. However, in practice loudspeakers that closely match human speech directivity may not be

available, so the question arises of how sensitive a measurement would be to deviations from ideal directivity. In this paper we

consider the potential influence that sound source directivity has on STI in terms of source-receiver distance, room volume, room

reverberation time, and background noise level.

Previous studies by Bozzoli et al. (2005a, 2005b) and Stewart and Cabrera (2007) have explored the issue of directivity of STI sound

sources through physical measurement. By testing a variety of loudspeaker enclosures approximating the human form in various

room acoustical contexts, Bozzoli et al. (2005a) found that directivity appeared to have little effect on STI measurements in normal

rooms (classrooms), but did have a substantial effect in a car cabin. Bozzoli et al. (2005b) found that the measured average directivityof ten talkers was a good match for the directivity of a Bruel & Kjaer 4128C head and torso simulator. However, one issue with this

is that a head and torso simulator is very expensive and somewhat fragile, and so is not necessarily a practical solution to STI

measurement. Stewart and Cabrera (2007) examined the effect of mouth size on the directivity of a head and torso simulator, finding

a substantial effect at some frequencies. When the effect of mouth size on STI measurements in real rooms (a lecture theatre and ameeting room) was tested, there was little effect, with only a modest variation in STI in one of the positions tested.

These measurement-based studies are limited by the effort involved in making physical measurements, and so only offer a glimpse of

the range of possible scenarios. Hence the present paper takes a theoretical appoach to the question.

MODEL OF STI

According to Houtgast et al. (1980), the modulation transfer function of a room acoustical system, m(F), can be estimated from

equation 1, based on principles from statistical room acoustics. The modulation transfer function is formed by a set of moduation

reduction coefficients (i.e., values that enumerate the extent to which the modulation depth of a modulated signal is reduced by theacoustic system) for a range of modulation frequencies (F), which for STI measurements span 0.63 Hz to 12.5 Hz in 1/3-octave

intervals. In STI measurements, this modulation transfer function is calculated for each of seven carrier frequency octave bands, from

125 Hz to 8 kHz.

m(F) =A2 + B2

C

A =QtQl

r2

+1

rc21 + 2!FT

13.8( )2"

#$%

24-26 November 2008, Geelong, Australia Proceedings of ACOUSTICS 2008

2 Acoustics 2008

B =2!FT

13.8rc

21 + 2!FT

13.8( )2"

#$%

C =QtQl

r2+1

rc2+Qt !10

LN "Lt ,1m( ) / 10(1)

Here Qt is the directivity factor of the talker, and Ql is the directivity factor of the listener. In emulating STI measurements Qt is the

directivity of the measurement loudspeaker and Ql=1 (an omidirectional measurement microphone is used). For a sound source,directivity factor is the ratio of sound intensity radiated in the reference direction to the average sound intensity radiated in all direc-

tions at a given distance from the source. Usually the on-axis direction (i.e. the normal to the face of the sound source) is taken as the

reference direction. Alternatively, directivity index, DI, is directivity factor expressed in decibels, i.e.

QDI log10= (2)

Returning to equation 1, the source-receiver distance is r. The critical distance (the distance from a sustained omnidirectional source

for which the direct and reverberant fields contribute equally to intensity) is denoted rc. T is the reverberation time of the room at the

carrier frequency under consideration. LN is the background noise level, and Lt, 1m is the long term equivalent sound pressure level of

the speaker at a distance of 1 m. This level is standardised in IEC60268-16(2003) as total of 60 dB(A) summed across the seven (formale speech) or six (for female speech) octave bands.

Each of these modulation reduction coefficients is converted to an ‘apparent signal-to-noise ratio’, (S/N)app,F,oct, stated in decibels

(equation 3). The subscript ‘oct’ refers to the fact that this calculation would normally be performed for each octave band carrier

frequency (as well as each modulation frequency, F). This value is influenced by both the actual noise level and the temporal smear-ing caused by reverberation. It is determined in the following manner:

S / N( )app, F,oct =10 logm F( )oct

1 !m F( )oct(3)

These values are then clipped, so that values greater than 15 dB are reduced to 15 dB, and values less than -15 dB are increased to -

15 dB.

The mean apparent signal-to-noise ratio for each carrier octave band is calculated according to equation 4:

S / N( )app,oct =S / N( )

app, F, oct!nF

(4)

Here, nF is the number of values (modulation frequencies) represented in a carrier octave band, which is equal to 14 for STI meas-urements (i.e., the 14 frequencies from 0.63 Hz to 12.5 Hz).

The apparent signal-to-noise ratios (one for each of the seven octave band carrier frequencies in STI measurements), are converted to

sound transmission indices according to equation 5.

STIoct =S / N( )

app,oct+ 15

30(5)

Although a full calculation of STI would then involve a weighted combination of seven STIoct values, taking into account interactions

between the bands due to masking, for the modelling in this paper we omit these final steps. This allows us to consider just one re-

verberation time and background noise level for a given output value (instead of reverberation time and noise varying between octavebands), and so provides a straightforward way of modelling the effect of directivity on STI. In the following discussion of our model-

ling, we use the term ‘STI’ to refer to what would be a STIoct value in a full calculation.


2 Acoustics 2008

B =2!FT

13.8rc

21 + 2!FT

13.8( )2"

#$%

C =QtQl

r2+1

rc2+Qt !10

LN "Lt ,1m( ) / 10(1)





QDI log10= (2)









1 !m F( )oct(3)


15 dB.



app, F, oct!nF

(4)




STIoct =S / N( )

app,oct+ 15

30(5)





(1)

HereQtisthedirectivityfactorofthetalker,andQlisthedirectivityfactorofthelistener.InemulatingSTImeasurementsQtisthedirectivityofthemeasurementloudspeakerandQl=1(anomidirectionalmeasurementmicrophone isused). Forasoundsource,directivityfactoristheratioofsoundintensityradiatedinthereferencedirectiontotheaveragesoundintensityradiated inalldirectionsatagivendistance fromthesource.Usuallytheon-axisdirection(i.e.,thenormaltothefaceofthesoundsource)istakenasthereferencedirection.Alternatively,directivityindex,DI,isdirectivityfactorexpressedindecibels,i.e.,


2 Acoustics 2008

B =2!FT

13.8rc

21 + 2!FT

13.8( )2"

#$%

C =QtQl

r2+1

rc2+Qt !10

LN "Lt ,1m( ) / 10(1)





QDI log10= (2)









1 !m F( )oct(3)


15 dB.



app, F, oct!nF

(4)




STIoct =S / N( )

app,oct+ 15

30(5)





(2)

Returningtoequation1,thesource-receiverdistanceisr.Thecriticaldistance(thedistancefromasustainedomnidirectionalsource for which the direct and reverberant fields contributeequallytomagnitude)isdenotedrc.Tisthereverberationtimeoftheroomatthecarrierfrequencyunderconsideration.LNisthebackgroundnoiselevel,andLt, 1misthelongtermequivalentsoundpressure levelofthespeakeratadistanceof1m.Thislevel is standardised in IEC60268-16(2003) as total of 60dB(A)summedacross the seven(formale speech)or six (forfemalespeech)octavebands.

Eachofthesemodulationreductioncoefficientsisconvertedto an ‘apparent signal-to-noise ratio’, (S/N)app,F,oct, stated indecibels(equation3).Thesubscript‘oct’referstothefactthatthiscalculationwouldnormallybeperformedforeachoctavebandcarrierfrequency(aswellaseachmodulationfrequency,F).Thisvalueisinfluencedbyboththeactualnoiselevelandthetemporalsmearingcausedbyreverberation.Itisdeterminedinthefollowingmanner:


2 Acoustics 2008

B =2!FT

13.8rc

21 + 2!FT

13.8( )2"

#$%

C =QtQl

r2+1

rc2+Qt !10

LN "Lt ,1m( ) / 10(1)





QDI log10= (2)









1 !m F( )oct(3)


15 dB.



app, F, oct!nF

(4)




STIoct =S / N( )

app,oct+ 15

30(5)





(3)

These values are then clipped, so that values greater than15dBarereducedto15dB,andvalues lessthan-15dBareincreasedto-15dB.

The mean apparent signal-to-noise ratio for each carrieroctavebandiscalculatedaccordingtoequation4:


2 Acoustics 2008

B =2!FT

13.8rc

21 + 2!FT

13.8( )2"

#$%

C =QtQl

r2+1

rc2+Qt !10

LN "Lt ,1m( ) / 10(1)





QDI log10= (2)









1 !m F( )oct(3)


15 dB.



app, F, oct!nF

(4)




STIoct =S / N( )

app,oct+ 15

30(5)





(4)

Here,nFisthenumberofvalues(modulationfrequencies)representedinacarrieroctaveband,whichisequalto14forSTImeasurements (i.e., the14 frequencies from0.63Hz to12.5Hz).

Theapparentsignal-to-noiseratios(oneforeachofthesevenoctave band carrier frequencies in STI measurements), areconvertedtosoundtransmissionindicesaccordingtoequation5.


2 Acoustics 2008

B =2!FT

13.8rc

21 + 2!FT

13.8( )2"

#$%

C =QtQl

r2+1

rc2+Qt !10

LN "Lt ,1m( ) / 10(1)





QDI log10= (2)









1 !m F( )oct(3)


15 dB.



app, F, oct!nF

(4)




STIoct =S / N( )

app,oct+ 15

30(5)





(5)

Although a full calculation of STI would then involvea weighted combination of seven STIoct values, taking intoaccountinteractionsbetweenthebandsduetomasking,forthemodellinginthispaperweomitthesefinalsteps.Thisallowsustoconsiderjustonereverberationtimeandbackgroundnoiselevel for a given output value (instead of reverberation timeand noise varying between octave bands), and so provides astraightforward way of modelling the effect of directivity onSTI.Inthefollowingdiscussionofourmodelling,weusetheterm ‘STI’ to refer to what would be a STIoct value in a fullcalculation.

STIhasavaluebetween0and1,where1correspondstogoodtransmission.Dependingonthecontext,avalueof0.45(thelowerlimitof‘fair’)andhighermayrepresentacceptableintelligibility.

CalculatedSTIandSensitivitytoDirectivityFactor

In order to characterise the effect of source directivity onSTIwehaveobtainedvaluesfromthemodeldescribedintheprevious section for directivity factors (Qt) ranging between0.125 and 8 using logarithmic steps (or directivity indicesbetween-9and9dBusinglinearsteps).Ofcourseitwouldbemostunusualtointentionallyemployasoundsourcewitha directivity factor of less than 1 (meaning that the sourceis directional, but its energy is radiated predominantly in adirectionotherthanthereferencedirection–forexamplewhenaperson speakswith theirback facing the listener). On theotherhand, thereare situationswhere someflexibility in thetalker’sorientation isdesirable, so itmaybeuseful todesignagainst potential problems arising from directivity indices oflessthan1.Inalatersectionofthepaperweconsidertherangeofdirectivities(asafunctionoffrequency)forhumantalkersand two electroacoustic sound sources designed for speechemulation.

Dow

nloa

ded

by [

Flor

ida

Atla

ntic

Uni

vers

ity]

at 1

3:18

16

Nov

embe

r 20

14

73KenStewartandDensilCabrera DirectivityandSpeechTransmissionIndex

Weconsidertwocases:(i)TheeffectonSTIof changing sourcedirectivitywhilst

maintaining constant on-axis free field sound pressurelevelat1m;and

(ii)TheeffectonSTIofchangingsourcedirectivitywhilstmaintainingconstantsourcepower.

Thefirstcase isrelevanttoconsideringtheextenttowhichthe directivities of various potential sound sources for STImeasurement would influence the result. The second case isrelevanttoconsideringtheeffectofsourceorientationonSTI.

The standard deviation of STI for 0.125 ≤ Qt ≤ 8 wastaken as the indicator of sensitivity of STI to directivity. Analternative approach might have been to take the differencebetweenmaximumandminimum(whichwouldhaveyieldedlarger values), but we prefer standard deviation because it isderivedfromthefullsetofcalculatedvaluesratherthanthetwoextremes.

Fortherangeofdirectivityindicesindicatedabove,wethendeterminedSTI considering the followingparameters: source-receiver distance (r), reverberation time (T), room volume(whichtogetherwithreverberationtimeaffectscriticaldistancerc),andbackgroundnoiselevel(LN).Threeroomvolumeswere

considered–130m3,1300m3and13000m3.Valuesofrwerescaledbythecuberootofroomvolume(0.05m–6.4minthe130m3room,0.11m–13.79minthe1300m3room,and0.23m–29.71minthe13000m3room).Valuesofreverberationtime ranged between 0.0625 s and 8 s (not scaled by roomvolume). Three background noise levels were considered:negligiblenoise,-15dBrelativetoanechoicspeechat1m(whichcouldbeinterpretedasbeingequivalentto45dBA),and-5dBrelativetoanechoicspeechat1m(i.e.,equivalentto55dBA).

MeanvaluesofSTIfortherangeofdirectivityfactorsareshownin Figure 1 in terms of source-receiver distance, reverberationtime,roomvolumeandbackgroundnoiselevel.Almostidenticalmeanvaluesarefoundforthetwocases(constanton-axis freefieldpressure,andconstantpower),soonlythefirstofthesecasesispresentedhere.Figures2and3showstandarddeviationsofSTIduetovariationofsourcedirectivityfactor,foreachofthetwocases.ThevaluesofFigure1aremainlyprovided tohelpinterpretFigures2and3.

Effect of Directivity in Low Background NoiseResultsforthenegligiblebackgroundnoiseconditionare

shown incolumn1ofFigures1-3. Aswouldbe expected,


4 Acoustics 2008

Mean values of STI for the range of directivity factors are shown in Figure 1 in terms of source-receiver distance, reverberation time,

room volume and background noise level. Almost identical mean values are found for the two cases (constant on-axis free fieldpressure, and constant power), so only the first of these cases is presented here. Figures 2 and 3 show standard deviations of STI due

to variation of source directivity factor, for each of the two cases. The values of Figure 1 are mainly provided to help interpret

Figures 2 and 3.

Effect of directivity in low background noise

Results for the negligible background noise condition are shown in column 1 of Figures 1-3. As would be expected, a long reverbera-

tion time and large source-receiver distance yields low STI values (Figure 1, column 1). There is no difference in the effect of direc-

tivity on STI between the two source conditions (constant pressure in Figure 2 versus constant power in Figure 3). Those sub-charts

show that sensitivity to directivity reaches a maximum at relatively short source-receiver distances, and that this peak sensitivity in-creases with reverberation time. The source-receiver distance for peak sensitivity decreases with increasing reverberation time, as

might be expected from the reduction in critical distance. However peak sensitivity is not at the critical distance, but instead is around

60% of the critical distance (however, this should not be taken as a general principle because it depends on the range of directivity

factors evaluated). Scaling the source-receiver distance by the cube root of room volume results in a similar pattern of sensitivitiesfor the three room volumes evaluated.

Effect of background noise on directivity sensitivity

Introducing background noise makes STI less dependent on reverberation time, especially in the larger volume rooms (Figure 1, col-

umns 2 and 3). In the case of constant source pressure (Figure 2) background noise introduces a second peak area of sensitivity tosource directivity, at large source-receiver distances. This area of peak sensitivity moves from short to longer reverberation times as

Figure 1. Mean STI for a range of source directivity factors (Qt) from 0.125 to 8. Results are shown as a function of reverbera-

tion time, source-receiver distance, room volume and background noise relative to anechoic on-axis speech at 1 m. Whilevalues are for the case of constant on-axis pressure, almost identical values are found for the case of constant source power.

Figure 1: Mean STI for a range of source directivity factors (Qt) from 0.125 to 8. Results are shown as a function of reverberation time, source-receiver distance, room volume and background noise relative to anechoic on-axis speech at 1 m. While values are for the

case of constant on-axis pressure, almost identical values are found for the case of constant source power.

Dow

nloa

ded

by [

Flor

ida

Atla

ntic

Uni

vers

ity]

at 1

3:18

16

Nov

embe

r 20

14


a long reverberation time and large source-receiverdistanceyields low STI values (Figure 1, column 1). There is nodifferenceintheeffectofdirectivityonSTIbetweenthetwosourceconditions(constantpressureinFigure2versusconstantpower in Figure 3). Those sub-charts show that sensitivitytodirectivity reaches amaximumat relatively short source-receiver distances, and that this peak sensitivity increaseswith reverberation time. The source-receiver distance forpeaksensitivitydecreaseswithincreasingreverberationtime,asmightbeexpectedfromthereductionincriticaldistance.However peak sensitivity is not at the critical distance, butinsteadisaround60%ofthecriticaldistance(however,thisshouldnotbetakenasageneralprinciplebecauseitdependson the range of directivity factors evaluated). Scaling thesource-receiver distance by the cube root of room volumeresultsinasimilarpatternofsensitivitiesforthethreeroomvolumesevaluated.

Effect of Background Noise on Directivity SensitivityIntroducing background noise makes STI less dependent

on reverberation time, especially in the larger volume rooms

(Figure1,columns2and3). In thecaseofconstant sourcepressure (Figure 2) background noise introduces a secondpeak area of sensitivity to source directivity, at large source-receiver distances. This area of peak sensitivity moves fromshort to longer reverberation times as the background noiselevelincreases,andastheroomvolumeincreases.Hencethisappearsasaninteractioneffectbetweenbackgroundnoiseandreverberation,ratherthananeffectofnoisealone.Thereasonfor this second peak is that in varying the directivity factor,themodelismaintainingtheon-axisanechoicsoundpressurelevel–hencethepowerofthesourceincreasesasthedirectivitydecreases.Inthesecondpeakarea,itisthisconsequentchangeinthepowerofthesource,ratherthandirectlyitsdirectivity,that is contributing to higher STI values for a given roomacousticalandbackgroundnoisecondition.Inthelargeroomconditions, the secondpeak areaoccurs in areasof relativelylowSTI (considerFigure2 in relation toFigure1), in somecases much lower than a practically useful value for speechcommunication.InthesmallroomSTIvaluesarereasonablygoodinthesecondpeakarea,andsoitmaybecomerelevanttounderstandingsmallerspaces.


Acoustics 2008 5

the background noise level increases, and as the room volume increases. Hence this appears as an interaction effect between back-

ground noise and reverberation, rather than an effect of noise alone. The reason for this second peak is that in varying the directivityfactor, the model is maintaining the on-axis anechoic sound

pressure level – hence the power of the source increases as

the directivity decreases. In the second peak area, it is this

consequent change in the power of the source, rather thandirectly its directivity, that is contributing to higher STI

values for a given room acoustical and background noise

condition. In the large room conditions, the second peak area

occurs in areas of relatively low STI (consider Figure 2 inrelation to Figure 1), in some cases much lower than a practi-

cally useful value for speech communication. In the small

room STI values are reasonably good in the second peak

area, and so it may become relevant to understanding smallerspaces.

In the case of maintaining constant source power, there is no

second peak area of sensitivity, and the introduction of back-

ground noise increases the sensitivity of STI to directivity atshort reverberation times (Figure 3).

Floor and ceiling effects are important in influencing the dark

areas (low standard deviations) of Figures 2 and 3. In condi-

tions of very high mean STI, or very low mean STI there is

little scope for directivity to influence STI. However, peak

sensitivity is not generally at a mean STI of 0.5, but at 0.7--0.8 in rooms with negligible noise.

An important consideration here is that the range of directiv-

ity factors that were evaluated (Qt of 0.125 to 8) influences

both the mean and the standard deviation, and for some pur-poses, evaluating directivity factors of less than 1 is much

less relevant than higher directivity factors – especially in the

first case (constant on-axis pressure), which might be inter-

preted as being relevant to the selection of loudspeakers forSTI evaluation. If larger Qt values are evaluated, then the

peak sensitivity is shifted to greater distances, and potentially

beyond the critical distance. This point is considered further

towards the end of the paper.

Figure 2. Standard deviation of STI for a range of source directivity factors (Qt) from 0.125 to 8, for the case of constant on-

axis sound pressure. Results are shown as a function of reverberation time, source-receiver distance, room volume and back-

ground noise relative to anechoic on-axis speech at 1 m.Figure 2: Standard deviation of STI for a range of source directivity factors (Qt) from 0.125 to 8, for the case of constant on-axis sound pressure. Results are shown as a function of reverberation time, source-receiver distance, room volume and background noise

relative to anechoic on-axis speech at 1 m.

Dow

nloa

ded

by [

Flor

ida

Atla

ntic

Uni

vers

ity]

at 1

3:18

16

Nov

embe

r 20

14


In the case of maintaining constant source power, thereisnosecondpeakareaof sensitivity,andthe introductionofbackgroundnoiseincreasesthesensitivityofSTItodirectivityatshortreverberationtimes(Figure3).

Floor and ceiling effects are important in influencing thedark areas (low standard deviations) of Figures 2 and 3. InconditionsofveryhighmeanSTI,orverylowmeanSTIthereislittlescopefordirectivitytoinfluenceSTI.However,peaksensitivityisnotgenerallyatameanSTIof0.5,butat0.7--0.8inroomswithnegligiblenoise.

An important consideration here is that the range ofdirectivity factors that were evaluated (Qt of 0.125 to 8)influencesboththemeanandthestandarddeviation,andforsomepurposes, evaluatingdirectivity factorsof less than1 ismuch less relevant thanhigherdirectivity factors–especiallyin the first case (constant on-axis pressure), which might beinterpretedasbeingrelevanttotheselectionofloudspeakersforSTIevaluation.Iflarger Qtvaluesareevaluated,thenthepeaksensitivityisshiftedtogreaterdistances,andpotentiallybeyondthecriticaldistance.Thispointisconsideredfurthertowardstheendofthepaper.

DirectivityIndicesofSpeechSourcesOn-Axis

The interpretation of the results of this model requiressomeknowledgeofthedirectivityindicesofspeechsources.Inthissectionweconsidertheon-axisdirectivitiesofpeople,aswellastwoelectroacousticsourcesthatmightbeusedforSTImeasurement.

The most extensive measurements available for humanspeechdirectivitycomefromastudybyChuandWarnock,(2002). For40subjects, theymeasured long termaveragesound pressure levels of conversational speech in 105directions around the talker (with 9 elevation anglesand 13 azimuth angles). They also performed the samemeasurementswithaBruelandKjaertype4128Cheadandtorsosimulator(Figure4).Tousetheirresultsinthisstudy,wehavecollapsedtheirmeasurementsintodirectivityfactorvalues.Resultsforon-axisdirectivitiesareshowninFigure5.

Forthisstudywealsomeasuredthedirectivityofaloud-speakerdesignedforatypeofSTImeasurement–theNTITalkbox (Figure 4). This loudspeaker is part of a kit for


6 Acoustics 2008

DIRECTIVITY INDICES OF SPEECH SOURCES On-axis

The interpretation of the results of this model requires some knowledge of the directivity indices of speech sources. In this section we

consider the on-axis directivities of people, as well as two electroacoustic sources that might be used for STI measurement.

The most extensive measurements available for human speech directivity come from a study by Chu and Warnock (2002). For 40

subjects, they measured long term average sound pressure levels of conversational speech in 105 directions around the talker (with 9elevation angles and 13 azimuth angles). They also performed the same measurements with a Bruel and Kjaer type 4128C head and

torso simulator. To use their results in this study, we have collapsed their measurements into directivity factor values. Results for on-

axis directivities are shown in Figure 5.

For this study we also measured the directivity of a loudspeaker designed for a type of STI measurement – the NTI Talkbox. Thisloudspeaker is part of a kit for the direct measurement of STIPA (a simplified version of STI intended particularly for measuring

public address systems).

In the context of a public address system, the measurement loudspeaker would be put at a person’s (e.g., announcer’s) position with

the same relationship to the system’s microphone as the real person – thereby factoring in the acoustic conditions of an announce-ment booth for example. Nevertheless, this kit is also a convenient tool for measuring STIPA in purely acoustical contexts, and in-

deed may provide more robust measurements than RASTI (Room Acoustics Speech Transmission Index) in environments where

reverberation time varies significantly across the 125 Hz – 8 kHz frequency range. This kit is smaller, cheaper and more rugged than

a Bruel & Kjaer head and torso simulator with an artificial mouth.

Figure 3. Standard deviation of STI for a range of source directivity factors (Qt) from 0.125 to 8, for the case of constant

power. Results are shown as a function of reverberation time, source-receiver distance, room volume and background noiserelative to anechoic speech at 1 m when Qt = 1.

Figure 3: Standard deviation of STI for a range of source directivity factors (Qt) from 0.125 to 8, for the case of constant power. Results are shown as a function of reverberation time, source-receiver distance, room volume and background noise relative to anechoic

speech at 1 m when Qt = 1.

Dow

nloa

ded

by [

Flor

ida

Atla

ntic

Uni

vers

ity]

at 1

3:18

16

Nov

embe

r 20

14


the direct measurement of STIPA (a simplified version ofSTI intended particularly for measuring public addresssystems).

Inthecontextofapublicaddresssystem,themeasurementloudspeaker would be put at a person’s (e.g., announcer’s)positionwiththesamerelationshiptothesystem’smicrophoneastherealperson–therebyfactoringintheacousticconditionsof an announcement booth for example. Nevertheless,this kit is also a convenient tool for measuring STIPA in

purely acoustical contexts, and indeed may provide morerobustmeasurements thanRASTI (RoomAcousticsSpeechTransmission Index) in environments where reverberationtimevariessignificantlyacrossthe125Hz–8kHzfrequencyrange. Thiskit is smaller, cheaperandmore rugged thanaBruel andKjaer head and torso simulatorwith an artificialmouth.

Measurementsweremadeinananechoicroomatadistanceof 1.5 m. The loudspeaker was rotated using a turntable,

Figure 4: Photograph of an NTI Talkbox (left) and a Bruel and Kjaer Type 4128C head and torso simulator (right).


Acoustics 2008 7

Figure 4. NTI Talkbox directivity measured using Bruel &

Kjaer Electro-acoustic Pulse System Type 7907 with turn-

table in Anechoic Chamber

Measurements were made in an anechoic room at a distance of 1.5 m. The loudspeaker was rotated using a turntable, with measure-

ments made every 10 degrees. Microphone elevation angles were at 20 degree intervals, ranging from -40 degrees from the horizon-

tal to 80 degrees (90 degrees was also measured). Measurements were made using the Bruel & Kjaer steady state response Electro-acoustic Pulse System Type 7907. A photograph from the measurement is shown in Figure 4, and results for on-axis directivity are

shown in Figure 5.

Following Beranek (1986), the directivity of simple circular piston sources can be predicted based on the product of wave number

and their radius (which is equivalent to the ratio of circumference to wavelength). Values for an 80 mm diameter piston on the end ofa long cylinder are shown by way of comparison to the measured values of the NTI Talkbox on Figure 5 (the Talkbox has a loud-

speaker diaphragm diameter of about 80 mm). This shows greater directivity for the NTI Talkbox over much of the frequency range,

presumably because of the physical effect of the rectangular loudspeaker enclosure. In the high frequency range it is usual for the

loudspeaker cone to mainly radiate from the centre, reducing its effective diameter, which would counteract the added directivity dueto the box – and this is seen in the high frequency results.

Figure 5. On-axis directivities of human talkers and a Bruel & Kjaer 4128C head and torso simulator, derived from Chu and War-nock’s (2002) data; and the measured on-axis directivity of an NTI Talkbox, compared with the expected value for an 80 mm circular

piston on the end of a long cylinder.

Based on the values in Figure 5, we can expect directivity index values between about 0 dB and 5 dB for real people, or -1 dB and 6

dB for a 4128C head and torso simulator. When a more conventional loudspeaker is used, directivity is likely to be greater in the highfrequency range, with the NTI Talkbox attaining a directivity index greater that 12 dB in the very high frequency range. In the vi-

cinity of 2 kHz (the part of the spectrum to which speech intelligibility is particularly sensitive) the NTI Talkbox had a directivity

index approximately 3 dB greater (i.e. double directivity factor) than the humans and 4128C.

Effect of source rotation on directivity index

Directivity index can be calculated for any reference angle, and in this section we consider the directivities that occur as a source

rotates around the horizontal plane. Figure 6 shows the directivity indices for Chu and Warnock’s human data and the head and torso

simulator, with angle of rotation on the vertical axis. This shows that, in the high frequency range, there can be substantially negative

directivity indices, which our modelling of STI standard deviation suggests can be important in some circumstances for speech intel-ligibility.

Figure 5: On-axis directivities of human talkers and a Bruel and Kjaer 4128C head and torso simulator, derived from Chu and Warnock’s, (2002) data; and

the measured on-axis directivity of an NTI Talkbox, compared with the expected value for an 80 mm circular piston on the end of a long cylinder.

Dow

nloa

ded

by [

Flor

ida

Atla

ntic

Uni

vers

ity]

at 1

3:18

16

Nov

embe

r 20

14


with measurements made every 10 degrees. Microphoneelevationangleswereat20degreeintervals,rangingfrom-40degrees from the horizontal to 80 degrees (90 degrees wasalso measured). Measurements were made using the BruelandKjaersteadystateresponseElectro-acousticPulseSystemType7907.Resultsforon-axisdirectivityareshowninFigure5.

FollowingBeranek,(1986),thedirectivityofsimplecircularpistonsourcescanbepredictedbasedontheproductofwavenumberandtheirradius(whichisequivalenttotheratioofcircumferencetowavelength).Valuesforan80mmdiameterpiston on the end of a long cylinder are shown by way ofcomparison to themeasured values of theNTITalkboxonFigure5(theTalkboxhasaloudspeakerdiaphragmdiameterofabout80mm).ThisshowsgreaterdirectivityfortheNTITalkboxovermuchofthefrequencyrange,presumablybecauseofthephysicaleffectoftherectangularloudspeakerenclosure.In the high frequency range it is usual for the loudspeakerconetomainlyradiatefromthecentre,reducingitseffectivediameter,whichwouldcounteracttheaddeddirectivityduetothebox–andthisisseeninthehighfrequencyresults.

BasedonthevaluesinFigure5,wecanexpectdirectivityindexvaluesbetweenabout0dBand5dB for realpeople,or -1 dB and 6 dB for a 4128C head and torso simulator.When a more conventional loudspeaker is used, directivityis likely tobegreater in thehigh frequencyrange,with theNTITalkboxattainingadirectivityindexgreaterthat12dBin theveryhigh frequency range. In thevicinityof2kHz(the part of the spectrum to which speech intelligibility isparticularlysensitive)theNTITalkboxhadadirectivityindexapproximately 3 dB greater (i.e., double directivity factor)thanthehumansand4128C.

Effect of Source Rotation on Directivity IndexDirectivityindexcanbecalculatedforanyreferenceangle,

andinthissectionweconsiderthedirectivitiesthatoccurasasourcerotatesaroundthehorizontalplane.Figure6showsthedirectivityindicesforChuandWarnock’shumandataandtheheadandtorsosimulator,withangleofrotationontheverticalaxis.Thisshowsthat,inthehighfrequencyrange,therecanbesubstantiallynegativedirectivityindices,whichourmodellingofSTIstandarddeviationsuggestscanbeimportantinsomecircumstancesforspeechintelligibility.

Figure 7 shows directivity index for rotations of the NTITalkbox.Aswouldbeexpected,therearemuchmoresubstantialeffectsondirectivityindexastheloudspeakerrotates,especiallyinthehighfrequencyrange.

DiscussionTointerprettheseresultsintermsofthestandarddeviations

inthefirstpartofthepaper,weshouldconsidertherangeofvaluesusedtogeneratethestandarddeviations(-9dBto9dB).VariationsinSTIoflessthan0.05arelikelytobetoosmalltobeofconsequence. However, the largest standarddeviationsareoftheorderof0.2,which,ofcourserepresentawiderrangeofSTIvalues(thecorrespondingrangebetweenmaximumandminimumis0.57insomecases).Therangeofdirectivityindexvaluesseenasloudspeakersarerotatedcanextendsubstantiallybeyondthe-9dBto9dBrange,andsocanhavealargeeffect

Figure 6: Measured directivity index of people speaking, and of a head and torso simulator (HATS) as a function of frequency

and angle of rotation (based on Chu & Warnock, 2002).


8 Acoustics 2008

Figure 6. Measured directivity index of people speaking,

and of a head and torso simulator (HATS) as a function offrequency and angle of rotation (based on Chu and Warnock,

2002).

Figure 7 shows directivity index for rotations of the NTI Talkbox. As would be expected, there are much more substantial effects on

directivity index as the loudspeaker rotates, especially in the high frequency range.

Figure 7. Measured directivity index of an NTI Talkbox as a function of frequency and angle of rotation.

DISCUSSION

To interpret these results in terms of the standard deviations in the first part of the paper, we should consider the range of values used

to generate the standard deviations (-9 dB to 9 dB). Variations in STI of less than 0.05 are likely to be too small to be of conse-

quence. However, the largest standard deviations are of the order of 0.2, which, of course represent a wider range of STI values (thecorresponding range between maximum and minimum is 0.57 in some cases). However, the range of directivity index values seen as

loudspeakers are rotated can extend substantially beyond the -9 dB to 9 dB range, and so can have a large effect on octave band STI

under the right combination of acoustical conditions. Rotating a loudspeaker corresponds to the constant power case.

Figure 7: Measured directivity index of an NTI Talkbox as a function of frequency and angle of rotation.

Dow

nloa

ded

by [

Flor

ida

Atla

ntic

Uni

vers

ity]

at 1

3:18

16

Nov

embe

r 20

14


onoctavebandSTIundertherightcombinationofacousticalconditions.Rotatingaloudspeakercorrespondstotheconstantpowercase.

Ontheotherhand,inthecaseofconstanton-axispressure,it would be unlikely that negative directivity index valueswouldbeused in realisticapplications. Figure8 shows therelationshipbetweenthedistancethathaspeaksensitivitytodirectivity to thecriticaldistanceof theroomfor therangeofdirectivity factors evaluated forFigures1-3, and also fora range of directivity factors between 1 and 8 (noise-freeconditions). Note that these relationships are independentof room volume, and the same for constant pressure andconstantpowercases.

ConclusionsProceedings of ACOUSTICS 2008 24-26 November 2008, Geelong, Australia

Acoustics 2008 9

0

0.2

0.4

0.6

0.8

1

1.2

1.4

0.0

63

0.1

25

0.2

5

0.5 1 2 4 8

Reverberation Time (s)

Ra

tio

of

pe

ak

se

nsitiv

ity

tocritica

l

dis

tan

ce

Q=0.125-8

Q=1-8

Figure 8. The distance of peak sensitivity to source directivity divided by the critical distance as a function of reverberation time.Two ranges of directivities are evaluated: Qt = 0.125 – 8; and Qt = 1-8.

On the other hand, in the case of constant on-axis pressure, it would be unlikely that negative directivity index values would be used

in realistic applications. Figure 8 shows the relationship between the distance that has peak sensitivity to directivity to the critical

distance of the room for the range of directivity factors evaluated for Figures 1-3, and also for a range of directivity factors between 1and 8 (noise-free conditions). Note that these relationships are independent of room volume, and the same for constant pressure and

constant power cases.

CONCLUSIONS

Previous studies have found that source directivity often has little effect on STI measurements, but can have a substantial effects in

some circumstances. This paper has looked at the acoustical conditions for which source directivity has maximum influence on

speech transmission index. Maximum sensitivity is related to the critical distance and reverberation time in noise-free conditions. A

second area of sensitivity is seen when noise is introduced (assuming on-axis pressure remains constant while directivity changes).The applications for this work include prioritising issues in the design and selection of artificial speakers for STI measurement, as

well as understanding speech intelligibility over short source-receiver distances (for example, in meeting rooms or restaurants).

REFERENCESBeranek, L.L., 1986, Acoustics, American Institute of Physics.

Bozzoli, F., Bilzi, P. and Farina, A. 2005a, “Influence of artificial mouth’s directivity in determining speech

transmission index,” 119th Audio Eng. Soc. Conv., New York, USABozzoli, F., Viktrorovitch, M. and Farina, A. 2005b “Balloons of directivity of real and artificial mouth used in determining speech

transmission index,” 118th Audio Eng. Soc. Conv., Barcelona, Spain

Chu, W.T. and Warnock, A.C.C. 2002, Detailed Directivity of Sound Fields around Human Talkers, National Research Council

Canada IRC-RR-104Houtgast, T., Steeneken, H. and Plomp, R. 1980, “Predicting Speech Intelligibility in Rooms from the Modulation Transfer

Function,” Acta Acustica united with Acustica 46, 60-81

International Electrotechnical Commission, 2003, Sound system equipment – Part 16: Objective rating of speech intelligibility byspeech transmission index, IEC 60268-16

Stewart, K. and Cabrera, D. 2007, “Effect of artificial mouth size on speech transmission index,” 14th International Congress on

Sound and Vibration, Cairns, Australia

Figure 8: The distance of peak sensitivity to source directivity divided by the critical distance as a function of reverberation

time. Two ranges of directivities are evaluated: Qt = 0.125 – 8; and Qt = 1-8.

PreviousstudieshavefoundthatsourcedirectivityoftenhaslittleeffectonSTImeasurements,butcanhavea substantialeffects in some circumstances. This paper has looked at theacousticalconditionsforwhichsourcedirectivityhasmaximuminfluenceonspeechtransmissionindex.Maximumsensitivityis related to the critical distance and reverberation time innoise-freeconditions.Asecondareaofsensitivityisseenwhennoiseisintroduced(assumingon-axispressureremainsconstantwhile directivity changes). The applications for this workincludeprioritisingissuesinthedesignandselectionofartificialspeakersforSTImeasurement,aswellasunderstandingspeechintelligibilityovershortsource-receiverdistances(forexample,inmeetingroomsorrestaurants).

ReferencesBeranek,L.L.,1986.Acoustics,AmericanInstituteofPhysics.Bozzoli,F.,Bilzi,P.&Farina,A.2005a.Influenceofartificialmouth’s

directivityindeterminingspeechtransmissionindex,119th Audio Eng. Soc. Conv.,NewYork,USA.

Bozzoli, F., Viktrorovitch, M., & Farina, A. 2005b. Balloons ofdirectivityofrealandartificialmouthusedindeterminingspeechtransmission index, 118th Audio Eng. Soc. Conv., Barcelona,Spain.

Chu,W.T.,&Warnock,A.C.C.2002.Detailed Directivity of Sound Fields around Human Talkers,NationalResearchCouncilCanadaIRC-RR-104.

Houtgast,T.,Steeneken,H.,&Plomp,R.1980.PredictingSpeechIntelligibilityinRoomsfromtheModulationTransferFunction,Acta Acustica united with Acustica46,60-81.

International Electrotechnical Commission, 2003. Sound system equipment – Part 16: Objective rating of speech intelligibility by speech transmission index,IEC60268-16.

Stewart,K.,&Cabrera,D.2007.Effectofartificialmouthsizeonspeech transmission index, 14th International Congress on Sound and Vibration,Cairns,Australia.

Dow

nloa

ded

by [

Flor

ida

Atla

ntic

Uni

vers

ity]

at 1

3:18

16

Nov

embe

r 20

14

effect of acoustic environment on the sensitivity of speech transmission index to source directivity

Documents