MISLEADINGANDCONFLICTINGCUESINHUMANSOUNDLOCALIZATION By EricJohnMacaulay ADISSERTATION Submittedto MichiganStateUniversity inpartialentoftherequirements forthedegreeof Physics|DoctorofPhilosophy 2015 ABSTRACT MISLEADINGANDCONFLICTINGCUESINHUMANSOUND LOCALIZATION By EricJohnMacaulay Humansoundlocalizationintheazimuthalplaneisprimarilycuedbytheinterauraltime (ITD)andtheinteraurallevel(ILD).Forsinetoneswithfrequencies greaterthanabout1500Hz,theILDistheonlysteady-statecueavailabletolisteners.In freeplanewavesincidentonalisteneraroundtheheadwhichresultsinan acousticalbrightspotontheoppositesideoftheheadasthesource.ThisresultsintheILD cuebeingnon-monotonicwithazimuth.Listeners'localizationresponsestostimuliinfree arehighlycorrelatedwiththismisleadingILDcue,aslistenerserrdramaticallyinlo- calizationofsourceswithlargeazimuths.Discriminationexperimentresultsalsothe misleadingnatureoftheILDcue.Trainingandfeedbackprovideveryminimalperformance improvementsforlisteners.However,theintroductionofeitheramplitudemodulation,or narrow-bandnoiseofthesamebandwidthhelpedsomelistenersgreatly,andotherstoa lesserextent.Thenon-monotoniccueremainsprominentfornoisewithbandwidthsaswide as4octaveswhencenteredaround1500kHz. Amplitudemodulationintroducesanenvelopeinterauraltime(EITD)asa localizationcueforlisteners.Sinusoidally-modulatedamplitudemodulationsignalsbecome alteredbyalistener'sanatomy.TheEITDcueisoftenmisleadingandincwiththe ILDcue.Nevertheless,thetwocuestogetherusuallyresultinaccuratelocalization.The qualityoftheamplitudemodulationintheseexperimentswasnotdegradedtlyand doesnotimpacttheweightoftheEITD. Thisworkisdedicatedtomyparents. iii ACKNOWLEDGMENTS Thisworkwouldnothavebeenpossiblewithoutthehelpofmanyindividualsandorga- nizations.Myadvisor,Dr.WilliamM.Hartmann,hasprovidedinvaluableguidancetome duringmygraduatestudies.Ithankhimforhispatienceandwisdom. Dr.BradRakerdhasbeenanexcellentmentorandcollaboratorofmine.Hehasalways beeneagertosharehistimeandinsight,andisapleasuretoworkwith. Ithanktheothermembersofmyguidancecommitteefortheiradviceandinterestinmy research:Dr.AlexBrown,Dr.DouglasBrungart,Dr.JonPumplin,andDr.StuartTessmer. Dr.ConstantineTrahiotisprovidedanimportantrecommendationinadiscussionabout accountingforin-eardistortionproductsforAMtones. Mr.ZachRyancontributedimportanttechnicalworkintheearlyphasesofthenon- monotonicinteraurallevelexperiment.Similarly,Mr.ThomasAndrewscon- tributedtotheearlydevelopmentoftheamplitudemodulationlocalizationexperiment. Formergraduatestudent,Dr.PeterXinyaZhang,graciouslyhelpedtointroducethe psychoacousticslaboratoriestome.Formergraduatestudent,Dr.NeilAaronson,shared importantwordsofadviceaboutgraduateschoolandresearchinpsychoacoustics. Ihavealsohadthepleasureofsharingspaceandgooddiscussionsaboutresearch withDr.LarisaDunaiandDr.TianshuQu. ThisworkwassupportedbytheNationalInstituteonDeafnessandOtherCommunica- tionDisorders(NIDCD)oftheNIHunderGrantNo.DC00181,andbytheAirForcece ofScienResearch(AFOSR)underGrantNo.11N002. iv TABLEOFCONTENTS LISTOFTABLES .................................... vii LISTOFFIGURES ................................... x KEYTOSYMBOLS .................................. xx KEYTOABBREVIATIONS ............................. xxii Chapter0Introduction ............................... 1 Chapter1Non-monotonicInterauralLevel ............ 3 1.1Introduction....................................3 1.2GeneralMethods.................................8 1.3KEMARRecordings...............................12 1.3.1KEMARResults.............................13 1.4Experiments,Results,andDiscussion......................14 1.4.1Experiment1:IdenationExperiment................14 1.4.1.1IdenExperimentResults...............17 1.4.2Experiment2:DiscriminationExperiment...............29 1.4.2.1DiscriminationExperimentResults..............30 1.4.3Experiment3:IdenationwithTrainingandFeedbackExperiment33 1.4.3.1IdenwithTrainingandFeedbackResults......34 1.4.4Experiment4:IdenationwithAmplitudeModulationExperiment43 1.4.4.1IdenwithAmplitudeModulationResults......43 1.4.5Experiment5:IdenationwithNarrow-bandNoiseExperiment..46 1.4.5.1IdenwithNarrow-bandNoiseResults........47 1.5Theofband-widthonnon-monotonicity.................49 1.5.0.2Theofband-widthonnon-monotonicityresults....50 1.6MinimumAudibleAngle.............................58 1.7Conclusion.....................................60 Chapter2AmplitudeModulationLocalization ................ 62 2.1Introduction....................................62 2.1.1Problem1:AMQuality.........................68 2.1.2Problem2:EITDGroupDelay.....................73 2.1.2.1Problem2:KEMARGroupDelay..............73 2.2Methods......................................77 2.2.1ExperimentalSetup............................77 2.2.2ExperimentalConditionsandProcedure................77 2.2.3Listeners..................................79 v 2.2.4ComputerAnalysisofSignals......................79 2.3ResultsandAnalysis...............................82 2.3.1PhysicalResults..............................82 2.3.1.1Problem1:AMQualityResults................82 2.3.1.2Problem2:EnvelopeITDResults...............85 2.3.1.3KEMARHeadphonesMeasurements.............86 2.3.2PerceptualResults............................88 2.3.2.1InterauralCuesandResponses................88 2.3.2.2SAMandsine...................114 2.3.2.3CompressiveCues.......................120 2.3.2.4SAMinterauralcuereliability.................125 2.3.3Discussion.................................125 2.3.4AMQuality(Problem1).........................133 2.3.4.1Changeinresponsevs.AMquality..............133 2.3.4.2TheofAMqualityoncuestrength...........139 2.3.4.3Responseerrorratiovs.AMquality.............159 2.3.4.4Discussion............................160 2.4Conclusion.....................................164 APPENDIX ........................................ 166 BIBLIOGRAPHY .................................... 174 vi LISTOFTABLES Table1.1:Independentearstrategysuccessrates,andcorrelationsbetweenlis- tenerresponsesandperfectresponses..................27 Table1.2:ThePearsonproduct-momentcorrelationcotsforvarioussets ofdataforeachlistener,aswellasthemeancotsacrossallof thelisteners.NTF1referstotheepassesoftheexperiment withnotrainingandfeedback.NTF2referstothesecondvepasses oftheexperimentwithnotrainingandfeedback.TF1standsforthe epassesoftheexperimentwithtrainingandfeedback.TF2 referstothesecondepassesoftheexperimentwithtrainingand feedback.Foreachofthesehalfexperiments|whichconsistofe passesthroughtheloudspeakerarray|themeanresponseforeach loudspeakerwasusedinthecorrelationcalculations..........34 Table1.3:ThePearsonproduct-momentcorrelationcotsforvarioussets ofdataforeachlistener,aswellasthemeancotsacrossallof thelisteners.Abbreviationsarethesameasthoseexplainedinthe captionforTable1.2.\Perfect"referstoahypotheticallistenerwho alwaysgivesthecorrectresponsetoastimulus.............40 Table1.4:Pearsonproduct-momentcorrelationcotsforalltenpasses throughtheloudspeakerarray.Eachlistener'scotsareshown forvarioussetsofdataaswellasthemeancocientsacrossallof thelisteners................................41 Table1.5:Pearsonproduct-momentcorrelationcotsforvariousexperi- ments.Thecotsarecalculatedforeachlistener,andthemean cotsareaveragedovertheelisteners.Equation(1.13)is usedtocalculatethecots,exceptthatforagivenlistenerand experiment,thedatausedinthecorrelationcalculationwerethe meanresponsesofthetenpassesthrougheachloudspeaker.\Pure Tone"referstoexperiment1,\AmplitudeModulation"referstoex- periment4,and\Perfect"referstoahypotheticallistenerwhoalways givesthecorrectresponsetoastimulus.................45 vii Table1.6:Pearsonproduct-momentcorrelationcotsforvariousexperi- ments.Thecotsarecalculatedforeachlistener,andthemean cotsareaveragedovertheelisteners.Equation(1.13)is usedtocalculatethecots,exceptthatforagivenlistenerand experiment,thedatausedinthecorrelationcalculationwerethe meanresponsesofthetenpassesthrougheachloudspeaker.\Noise" referstheexperimentwithnarrow-bandnoise,\PureTone"refersto experiment1,\AmplitudeModulation"referstoexperiment4,and \Perfect"referstoahypotheticallistenerwhoalwaysgivesthecorrect responsetoastimulus..........................47 Table2.1:Thepercentageofamplitudemodulation, m ,means > 100%.The meansarecalculatedacrossidenticalintervalconditions.Thedata arecombinedacrossalllistenersandazimuths.............84 Table2.2:Valuesof m , ,and m e calculatedfromrecordingsofKEMARmanikin (largepinna)wearingheadphones(SennheiserHD535).Thesignals were100%SAMtoneswithcarrierfrequenciesof2,3,and4kHz,and amodulationfrequencyof100Hz.Themonauralparameterswere calculatedusingmatchedltering....................87 Table2.3:Fisher'smethodp-values.Thismeta-analysisusesaone-sided ˜ 2 test with2 n degreesoffreedomonthesummationoverthelogarithmsof thet-testp-valuesfortheloudspeakers,where n isthenumberoft- testp-valuesinthesummation.Resultsareshowforeachfrequency andlisteneroverthe13loudspeakers,foreachfrequencyacrossall listenersandloudspeakers,andacrossallfrequencies,listeners, andloudspeakers.............................109 Table2.4:Thecorrelationcots, r ,forFigs.2.44{2.46andfourothertests. Thedataincludealllistenersandallfrequencies.Foreachloud- speaker,thedataarethechangeinmeanresponses(AM sine)vs. themeanindependentvariable.For m -leftand m -rightfoldedover 1,valuesof m< 1aretransformedtovaluesof m> 1whilekeeping j m 1 j thesame.............................138 Table2.5:Pearsonproduct-moments, r ,fortheresponseresidualsfrombest- linesvs.theAMqualitymetricshownbythecolorscaleinFigs. 2.47{2.61.Thecorrelationsarecalculatedforeachlistener,frequency, andAMqualitymetric..........................159 viii Table.1:MeasurementsforSAMtoneinthereverberationroomatamodula- tionfrequencyof100Hz.Foreachangle,locationanddistance( d ), quantitiesshownaretheAMintheleftear( m left ),theAMinthe rightear( m right ),theQFMintheleftear( left ),theQFMinthe rightear( right ),theenvelopeinterauraltimeerence(EITD)in s,theinterauralenvelopecoherence(E.Coh.),andtheinteraural level(ILD)indB.......................170 Table.2:SameasTable.1butforthereverberationroomat500Hz......171 Table.3:SameisTable.1butforthethe\laboratory"at100Hz........172 Table.4:SameisTable.1butforthethe\laboratory"at500Hz........173 ix LISTOFFIGURES Figure1.1:ILDvs.azimuthforthesphericalheadmodelwithaheadradius of8.75cm.Atfrequenciesgreaterthanabout1000Hz,thecurves becomeseriouslynon-monotonic.....................5 Figure1.2:Thelevelvs.azimuthfortheleftandrightearsrelativetothelevel thatwouldbeatthecenteroftheheadifnoheadwerepresent.The ILDisplottedvs.azimuth.Theyareshownforthesphericalhead modelwithafrequencyof1500Hzandaheadradiusof8.75cm.The ILDisshowntobenon-monotonicbecausethelevelattheleftear isnon-monotonic.Thisisbecausethelevelattheleftearincreases afterapproximately50 ,whichisduetotheacousticalbrightspot..6 Figure1.3:Relationshipbetweenthelistener'sorientationandthearrayofloud- speakers.The13loudspeakers,numbered0{12,span90 .Theangu- larspacingbetweenloudspeakersis7 : 5 ................9 Figure1.4:Levels,ILD,andIPDvs.azimuthforaKEMARmanikin.KEMAR's owninternalmicrophonesandelectronicswereusedfortheserecord- ings.Theinterauralgraphsarepolygons,wherethemean thestan- darddeviationareplottedforeachloudspeaker,andstraightlinesare drawnbetweenthepoints.However,thestandarddeviationsareso smallthatnopolygoncanbeseen.Therewere10passesthrough theloudspeakerarray.Sinceeachloudspeakerispresentedtwiceina singlepass,thereare20measurementsforeachsourcenumber....15 Figure1.5:Levels,ILD,andIPDvs.azimuthforaKEMARmanikin.Probe microphoneswereinsertedintotheKEMAR'searcanalsandwere usedtomaketheserecordings.Theinterauralgraphsarepolygons, wherethemean thestandarddeviationsareplottedforeachloud- speaker,andstraightlinesaredrawnbetweenthepoints.Therewere 10passesthroughtheloudspeakerarray.Sinceeachloudspeakeris presentedtwiceinasinglepass,thereare20measurementsforeach sourcenumber...............................16 x Figure1.6:ResultsfortheidenexperimentforlistenerB.Panel(a) showslistenerresponsesforeachtrial.Responsenumberisplotted vs.sourcenumberorazimuth.Panel(b)showsthemeanlevelforthe near(right)andfar(left)ears.Thelevelshavebeensetto0dBat0 . Errorbarsshow thestandarddeviationoverthe20measurements ateachangle.Inpanel(c),thedatapointsshowthemeanresponse numberforeachsourcenumber,wheretheerrorbarsshow the standarddeviationoverthe10responsesateachsourcenumber.The shadedpolygonrepresentsthemean standarddeviationoftheILD foreachsourcenumber.Thecorrelationcot(CC)between thetwosetsofdatais0.952forlistenerB.Panel(d)showsthesame listenerresponsedataaspanel(c)andalsoshowstheITDmeanand standarddeviation,whichisrepresentedbytheshadedpolygon.The correlationcot(CC)betweenthetwosetsofdatais0.512...18 Figure1.7:SameasforFig.1.6exceptforlistenerMandthecorrelationco cientsforpanels(c)and(d)are0.986and0.743,respectively.....19 Figure1.8:SameasforFig.1.6exceptforlistenerNandthecorrelationco cientsforpanels(c)and(d)are0.945and0.293,respectively.....20 Figure1.9:SameasforFig.1.6exceptforlistenerEandthecorrelationco cientsforpanels(c)and(d)are0.991and0.443,respectively.....21 Figure1.10:SameasforFig.1.6exceptforlistenerXandthecorrelationco cientsforpanels(c)and(d)are0.920and0.740,respectively.....22 Figure1.11:SpeakerResponsevs.ILD.Eachdatapointrepresentsaparticular loudspeakerforagivenlistener.Eachlistener'sspeakerresponseis plottedversusthemeanILDforagivenloudspeaker.Theerrorbars showplusandminusthestandarddeviationforspeakerresponse. ThestandarddeviationsforILDarenotshown,butaregenerallyon theorderofabout1dB.Thetwocurvesaregivenbyequation1 wheretheuppercurveistothesedatawithaparameterof5.95, andthelowercurveistoYost's1981experimentwithaparameter of8.....................................23 Figure1.12:DiscriminationExperiment.Thepredictedpercentcorrectandactual percentcorrectareplottedagainsteachotherforallelisteners. Thepercentcorrectwaspredictedbasedonthedatafromindividual listenersintheidenexperiment................31 xi Figure1.13:Thepercentofcorrectresponsesvs.theaverageenceinILD betweenthespeakerpairs.TheinILDistakentobe therightspeaker'sILDminustheleftspeaker'sILD.Thestandard deviationsinthearetypicallyontheorderof1dBorless..32 Figure1.14:LoudspeakerresponsenumbersforlistenerBduringthetrainingand feedbackexperiment.Therearetenresponsespersource.Thecircles representresponsesfortheepassesthroughthe13-loudspeaker array,andthetrianglesrepresentresponsesforthesecondepasses throughthe13-loudspeakerarray.Thedatapointshavebeenjogged slightlyofthegridlinessothattheymayallbevisible......35 Figure1.15:SameasFig.1.14exceptforlistenerE.................36 Figure1.16:SameasFig.1.14exceptforlistenerM.................37 Figure1.17:SameasFig.1.14exceptforlistenerN.................38 Figure1.18:SameasFig.1.14exceptforlistenerX.................39 Figure1.19:Eachlistener'smeanresponseforeachsourcenumberintheampli- tudemodulationexperiment......................44 Figure1.20:Eachlistener'smeanresponseforeachsourcenumberforthenarrow- bandnoiseexperiment..........................48 Figure1.21:KEMARlevelsforright,leftandILDwitha1500Hzsine-tonestim- ulus.Therecordingsweremadeusingprobemicrophones.Levelsare shownwithrespecttospeakerzero,andareadjustedfortherelative levelofthestimulusforeachloudspeakeratthecenterofthearray.51 Figure1.22:KEMARlevelsforright,leftandILDwitha1/3-octave-bandstim- uluscenteredlogarithmicallyaround1500Hz.Therecordingswere madeusingprobemicrophones.Levelsareshownwithrespectto speakerzero,andareadjustedfortherelativelevelofthestimulus foreachloudspeakeratthecenterofthearray.............52 Figure1.23:SameasFig.1.22butwitha1-octaveband..............53 Figure1.24:SameasFig.1.22butwitha2-octaveband..............54 Figure1.25:SameasFig.1.22butwitha3-octaveband..............55 Figure1.26:SameasFig.1.22butwitha4-octaveband..............56 xii Figure1.27:AverageminimumaudibleanglefromMills,1958.Theesetsof dataarefortazimuths......................59 Figure2.1:Spectrumof100%amplitudemodulation.Thecarrierfrequency, ! c , hasanamplitudeof C .Thesidebands,at ! c ! m and ! c + ! m ,have amplitudesof C= 2.Thephaseofthelowersidebandis ˚ c ˚ a and thephaseoftheuppersidebandis ˚ c + ˚ a ...............63 Figure2.2:SAMsignalwith100%AM.Thecarrierfrequencyis2000Hzandthe modulationfrequencyis100Hz.NoQFMispresentinthissignal. Inthisexample,thesignalsinthetwoearsareidentical.......69 Figure2.3:Leftandrightwaveformsofarbitraryunitscalculatedfromearcanal recordings.Thecarrierfrequencywas2000Hzandthemodulation frequencywas100Hz.Valuesfortheamplitudemodulation,QFM, andenvelopemodulationfractionareshownontherightby m , , and m e ,respectively...........................71 Figure2.4:Leftandrightwaveformsofarbitraryunitscalculatedfromearcanal recordings..Thecarrierfrequencywas3000Hzandthemodulation frequencywas100Hz.Valuesfortheamplitudemodulation,QFM, andenvelopemodulationfractionareshownontherightby m , , and m e ,respectively...........................72 Figure2.5:WrappedIPD( 180 to+180 )vs.carrierfrequency.Measurements weremadeintheanechoicroomwiththe1-marray.Theazimuth was60 .Thesignalsweresinetonesat10Hzintervals........74 Figure2.6:Left(blue)andright(red)waveformsandenvelopesofKEMAR recordingsofSAMtonesinfreeThesignalwasa2240Hz carrierwithamodulationfrequencyof40Hz.............75 Figure2.7:Left(blue)andright(red)waveformsandenvelopesofKEMAR recordingsofSAMtonesinfreeThesignalwasa2325Hz carrierwithamodulationfrequencyof40Hz.............76 Figure2.8:Amplitudemodulation, m ,meansforeachloudspeaker,frequency, andlistener.Histogramsfortheleftandrightearsaredisplayed separately.Eachhistogramcontainsatotalof195values.......84 Figure2.9:QFM, ,meansforeachloudspeaker,frequency,andlistener.His- togramsfortheleftandrightearsaredisplayedseparately.Each histogramcontainsatotalof195values.................85 xiii Figure2.10:HistogramofEnvelopeITDmeansforeachloudspeaker,frequency andlistener................................86 Figure2.11:Responsesandprobe-microphonemeasurementsforlistenerB,2kHz. Thesourcenumbersspantherightfrontquadrant.ThePearson product-moment(PPM)correlationcotisshownbetweenthe meanresponsesandtheinterauralcues.(a)Sinetone:Circlesindi- catemeanresponsesanderrorbarsaretwostandarddeviationsin overalllength.ThehatchedregionshowstheILD.Itiscenteredon themeanandistwostandarddeviationshigh.(b)SAMtone:Circles indicatemeanresponsesanderrorbarsaretwostandarddeviations inoveralllength.ThehatchedregionshowstheILD.Itiscenteredon themeanandistwostandarddeviationshigh.(c)SAMtone:Circles indicatemeanresponsesanderrorbarsaretwostandarddeviations inoveralllength.ThehatchedregionshowstheEITD.Itiscentered onthemeanandistwostandarddeviationshigh...........90 Figure2.12:SameasFig.2.11butforlistenerBat3kHz..............91 Figure2.13:SameasFig.2.11butforlistenerBat4kHz..............92 Figure2.14:SameasFig.2.11butforlistenerCat2kHz..............93 Figure2.15:SameasFig.2.11butforlistenerCat3kHz..............94 Figure2.16:SameasFig.2.11butforlistenerCat4kHz..............95 Figure2.17:SameasFig.2.11butforlistenerMat2kHz.............96 Figure2.18:SameasFig.2.11butforlistenerMat3kHz.............97 Figure2.19:SameasFig.2.11butforlistenerMat4kHz.............98 Figure2.20:SameasFig.2.11butforlistenerLat2kHz..............99 Figure2.21:SameasFig.2.11butforlistenerLat3kHz..............100 Figure2.22:SameasFig.2.11butforlistenerLat4kHz..............101 Figure2.23:SameasFig.2.11butforlistenerVat2kHz..............102 Figure2.24:SameasFig.2.11butforlistenerVat3kHz..............103 Figure2.25:SameasFig.2.11butforlistenerVat4kHz..............104 xiv Figure2.26:p-valuevs.speakernumberforchangesinresponseat2kHz.Each listenerisplottedaccordingtothesymbolsinthelegend.Thep- valuesarebasedonatwo-tailedt-testandrepresenttheprobability thattheresponsesinthesinetoneandAMtrialsarestatistically similar.Areferencelineisshownfor =0 : 05.............106 Figure2.27:SameasFig.2.26butfor3kHz.....................107 Figure2.28:SameasFig.2.26butfor4kHz.....................108 Figure2.29:Pearsonproduct-moment(PPM)correlationcotsforperfect responses(sourcenumber)andactualresponsesareshownforAM inredandsineinblue.Foreachlistener,thethreefrequenciesand themeanandstandarddeviationacrossfrequenciesareshown.In thelowerrightisthemeanandstandarddeviationacrossalllisteners foreachfrequencyandthemeanandstandarddeviationacrossall listenersandfrequencies.Theerrorbarsaretwostandarddeviations inoveralllength.............................111 Figure2.30:Pearsonproduct-moment(PPM)correlationcontsforAMre- sponsesandAMILDareinred.Correlationsforsineresponsesand sineILDareinblue.Foreachlistener,thethreefrequenciesandthe meanandstandarddeviationacrossfrequenciesareshown.Inthe lowerrightisthemeanandstandarddeviationacrossalllisteners foreachfrequencyandthemeanandstandarddeviationacrossall listenersandfrequencies.Theerrorbarsaretwostandarddeviations inoveralllength.............................112 Figure2.31:Pearsonproduct-moment(PPM)correlationcotsforAMILD andEITDareinred.CorrelationsforAMILDandsineILDare inblue.Foreachlistener,thethreefrequenciesandthemeanand standarddeviationacrossfrequenciesareshown.Inthelowerright isthemeanandstandarddeviationacrossalllistenersforeachfre- quencyandthemeanandstandarddeviationacrossalllistenersand frequencies.Theerrorbarsaretwostandarddeviationsinoverall length...................................113 Figure2.32:Pearsonproduct-moment(PPM)correlationcontsforAMre- sponsesandAMILDareinred.CorrelationsforAMresponsesand EITDareinblue.Foreachlistener,thethreefrequenciesandthe meanandstandarddeviationacrossfrequenciesareshown.Inthe lowerrightisthemeanandstandarddeviationacrossalllisteners foreachfrequencyandthemeanandstandarddeviationacrossall listenersandfrequencies.Theerrorbarsaretwostandarddeviations inoveralllength..............................115 xv Figure2.33:Pearsonproduct-moment(PPM)correlationcotsforthechange inresponseandthechangeinILDareshowninred.Correlationco- tsforthechangeinresponsesandtheEITDareshowninblue. Foreachlistener,thethreefrequenciesandthemeanandstandard deviationacrossfrequenciesareshown.Inthelowerrightisthemean andstandarddeviationacrossalllistenersforeachfrequencyandthe meanandstandarddeviationacrossalllistenersandfrequencies.The errorbarsaretwostandarddeviationsinoveralllength........116 Figure2.34:Changeinresponse(AM sine(mean))vs.EITDforallnegative EITD.Alllistenersandfrequenciesatcombined.Theverticalaxss indicatesAMresponsesforindividualtrialsminusthemeansine responseforthesameloudspeaker.Thehorizontalaxisindicates theEITDmeasuredintheindividualAMtrials.Thecorrelation cot| r |,slope,andy-interceptforthebestlineareabove theplots.ThecolorscaleindicatesthevalueofthechangeinILDbe- tweentheindividualAMandaveragesineruns.AfewoftheILD valuesclipthetopofthecolorscaleat+5dB.Areferencelinefor zeroresponseisshown.........................118 Figure2.35:ResidualsfromthebestlineinFig.2.34vs.ILD.Alineofbest isshownaswellasreferencelinesthroughtheorigin........119 Figure2.36:FIG.3fromYost(1981).Inthisheadphoneexperiment,listenerslat- eralizedon-goingIPDcuesat500Hz.Thetsymbolsindicate tlisteners,andtheverticallinesindicaterangesofresponses forIPDsof0 , 90 ,and180 .....................121 Figure2.37:ChangesinresponseandcompressedinterauralcuesforlistenerB. Therowis2kHz,thesecondrowis3kHz,andthethirdrowis 4kHz.Foreachplot,redcirclesindicatechangesinmeanresponse (AM sine)anderrorbarsaretwostandarddeviationsinoverall length.Forplots(a),(c),and(e),bluetrianglesindicatethechange incompressedILD(AM sine).Forplots(b),(d),and(f),bluedia- mondsindicatethecompressedEITD.Foreachplot,themaximized Pearsonproduct-moment(PPM)correlationcotbetweenthe changesinresponseandcompressedinterauralcueisshown,aswell asthecompressionexponent.......................126 Figure2.38:SameasFig.2.37butforlistenerC...................127 Figure2.39:SameasFig.2.37butforlistenerM...................128 Figure2.40:SameasFig.2.37butforlistenerL...................129 xvi Figure2.41:SameasFig.2.37butforlistenerV...................130 Figure2.42:MaximizedPearsonproduct-moment(PPM)correlationcots forthechangeinAMandsineresponsesandthechangeincompressed ILDareshowninred.CorrelationcotsforthechangeinAM andsineresponsesandthecompressedEITDareshowninblue.The greencirclesindicatethecompressionexponent.Foreachlistener, thethreefrequenciesandthemeanandstandarddeviationacross frequenciesareshown.Inthelowerrightisthemeanandstandard deviationacrossalllistenersforeachfrequencyandthemeanand standarddeviationacrossalllistenersandfrequencies.Theerror barsaretwostandarddeviationsinoveralllength...........131 Figure2.43:Pearsonproduct-moment(PPM)correlationcotsforAMtones. CorrelationsbetweenILDandsourceazimuthareinred.Correlations betweenEITDandsourceazimuthareinblue.Foreachlistener,the threefrequenciesandthemeanandstandarddeviationacrossfre- quenciesareshown.Inthelowerrightisthemeanandstandard deviationacrossalllistenersforeachfrequencyandthemeanand standarddeviationacrossalllistenersandfrequencies.Theerror barsaretwostandarddeviationsinoveralllength...........132 Figure2.44:Changeinresponsevs. m -leftear.Allelistenersareshownhere. Eachpointrepresentsthemeanchangeinresponseacross10trials vs.themean m across20intervalsforthelistener'sleft(far)ear.The correlationis r = 0 : 1031and r 2 =0 : 01063..............135 Figure2.45:Changeinresponsevs. m -rightear.Allelistenersareshownhere. Eachpointrepresentsthemeanchangeinresponseacross10trials vs.themean m across20intervalsforthelistener'sright(near)ear. Thecorrelationis r = 0 : 0973and r 2 =0 : 00947............136 Figure2.46:Changeinresponsevs.interauralenvelopecoherence.Allelis- tenersareshownhere.Eachpointrepresentsthemeanchangein responseacross10trialsvs.themean m across20intervals.The correlationis r = 0 : 0633and r 2 =0 : 00440..............137 xvii Figure2.47:Changesinresponsevs.EITDwithAMqualityforlistenerBat2 kHz.AllplotsdisplaythesamedataandintheAMquality metricshownincolor.TheverticalaxisindicatesAMresponsesfor individualtrialsminusthemeansineresponseforthesameloud- speaker.ThehorizontalaxisindicatestheEITDmeasuredinthe individualAMtrials.Thecorrelationcot| r |,slope,andy- interceptforthebestlineareabovetheplots.Thecolorscalein plot(a)indicatesthevalueof m intheleftearintheindividualAM trials.Thecolorscaleinplot(b)indicatesthevalueof m intheright earintheindividualAMtrials.Thecolorscaleinplot(c)indicates thevalueoftheenvelopecoherenceintheinindividualAMtrials. Thestandarddeviations, ˙ ,oftheAMqualityareshownabovethe plots....................................141 Figure2.48:SameasFig.2.47butforlistenerBat3kHz..............142 Figure2.49:SameasFig.2.47butforlistenerBat4kHz..............143 Figure2.50:SameasFig.2.47butforlistenerCat2kHz..............144 Figure2.51:SameasFig.2.47butforlistenerCat3kHz..............145 Figure2.52:SameasFig.2.47butforlistenerCat4kHz..............146 Figure2.53:SameasFig.2.47butforlistenerMat2kHz.............148 Figure2.54:SameasFig.2.47butforlistenerMat3kHz.............149 Figure2.55:SameasFig.2.47butforlistenerMat4kHz.............150 Figure2.56:SameasFig.2.47butforlistenerLat2kHz..............151 Figure2.57:SameasFig.2.47butforlistenerLat3kHz..............152 Figure2.58:SameasFig.2.47butforlistenerLat4kHz..............153 Figure2.59:SameasFig.2.47butforlistenerVat2kHz..............154 Figure2.60:SameasFig.2.47butforlistenerVat3kHz..............155 Figure2.61:SameasFig.2.47butforlistenerVat4kHz..............156 Figure2.62:ResidualsfromthebestlineinFig.2.47vs. m -left.Thecorrelation is r = 0 : 064and r 2 =0 : 0041......................158 xviii Figure2.63:TheratiooftheerrorinresponsermsforsinetonestoAMtones vs. m -leftaveragedacrossallspeakers.Thevelistenersandthree frequenciesareallrepresented.Thecorrelationis, r =0 : 1160and r 2 =0 : 01346................................161 Figure2.64:TheratiooftheerrorinresponsermsforsinetonestoAMtones vs. m -rightaveragedacrossallspeakers.Theelistenersandthree frequenciesareallrepresented.Thecorrelationis, r = 0 : 1459and r 2 =0 : 02129................................162 Figure2.65:TheratiooftheerrorinresponsermsforsinetonestoAMtones vs.theenvelopecoherenceaveragedacrossallspeakers.Thee listenersandthreefrequenciesareallrepresented.Thecorrelationis, r = 0 : 0221and r 2 =0 : 000488.....................163 xix KEYTOSYMBOLS A L :cosineintergraloftheleftear A R :cosineintergraloftherightear B L :sineintergraloftheleftear A R :sineintergraloftherightear x L :recordingintheleftear x R :recordingintherightear P L :\power"intheleftear P R :\power"intherightear ˚ L :phaseoftheleftear ˚ R :phaseoftherightear ILD :responseforagivenILD C :amplitudeofthecarrier m :amplitudemodulation ! m :modulationfrequency ˚ a :phaseoftheamplitudemodulation ˚ c :phaseofthecarrierfrequency ! :phaseexcursion ˚ f :phaseofthefrequencymodulation :quasifrequencymodulation ! ` :lowersideband ! u :uppersideband xx A c :cosineintegralofthecarrier B c :sineintegralofthecarrier A ` :cosineintegralofthelowersideband B ` :sineintegralofthelowersideband A u :cosineintegraloftheuppersideband B u :sineintegraloftheuppersideband E :envelope m e :envelopemodulationfraction xxi KEYTOABBREVIATIONS AM:amplitudemodulation ETID:envelopeinterauraltimee FM:frequencymodulation ILD:interaurallevel IPD:interauralphase ITD:interauraltimee KEMAR:KnowlesElectronicsManikinforAcousticResearch PPM:Pearsonproduct-moment SAM:sinusoidalamplitudemodulation QFM:quasifrequencymodulation xxii Chapter0 Introduction Soundlocalizationisaremarkablesensorycapabilityofhumansandotheranimals.Unlike vision(retina)ortouch(skin)wherelocalizationistopical,soundlocalizationneedstobe computed.Inhearing,topicalcodingcorrespondstofrequency.Comparedtosynapticdelays of1ms,theuniqueandremarkablecomputationalmachineryoftheauditorysystemfeatures interauraltime(ITD)sensitivityatleastassmallas20 .Soundlocalization asstudiedhereispartofthetopicofbinauralhearing,whichalsoincludesthetopicof binauraldetectionadvantage.Binauralhearinghasbeenstudiedfor100years.Itacquires newbecauseoftheincreasinguseofbilateralcochlearimplants. Therearetwoexperimentalstudiesonhumansoundlocalizationinthisthesis.Thenon- monotonicinteraurallevelstudyisinChapter1andtheamplitudemodulation localizationstudyisinChapter2.Ingeneral,thestudyofhumansoundlocalizationaims tounderstandwhylistenersperceivethelocationofsourcesofsoundwheretheydo.This isdoneinthreedimensions,andthecoordinatesystemofchoiceusestheazimuthalangle inthehorizontalplaneofthelistener'stwoears,theverticalangleasmeasuredabovethe horizontalplane,andthedistancefromthesourcetothelistener.However,thetwomost importantcuesforsoundlocalization,theinterauraltime(ITD)andtheinteraural level(ILD)areabletobestudiedbyonlyconsideringsoundsourcesthatarein thehorizontalplane.Thestudiesinthisthesisareeachrestrictedtosoundsourcesinthis plane. 1 Additionally,thesestudiesbothtakeplaceinfreeenvironmentinwhichthere arenoacousticalorreverberation.Theadvantageofthistypeofexperimentis thatthesoundeldcanbecontrolled.Alistenerispresentedwithastimulusfromonlyone directioninspace.Thelocationsofthesoundsourcesarealsofarenoughawayfromthe listenerthattoagoodapproximationthewavefrontsarrivingatthelocationofthelistener areplanewaves.Thelistener'staskistoreportonlytheazimuthalangle|notthedistance, andnottheelevation. Anotablefeatureinbothofthesestudiesistheextensiveuseofrecordingsofthesound pressurewavesinsidethelistener'searcanals.Theserecordingsarepossibleduetotheuseof smallprobemicrophoneswhichareinsertedintothelistener'searcanals.Theserecordings allowfortheinteraurallocalizationcues|theILDandandtheITD|tobecalculated. AlthoughtheILDandITDaretypicallyusefulcuesforlisteners,duetotheinteractionofthe incidentsoundwaveswithalistener'sanatomy,theycansometimeshavemisleadingqualities. Chapter1studiesasituationinwhichtheinteraurallevelcueismisleading,and Chapter2studiesasituationinwhichtheinterauraltimeintheenvelopeofa soundwaveismisleading,orinwiththetheILD. 2 Chapter1 Non-monotonicInterauralLevel 1.1Introduction Humansoundlocalizationreliesontwoprimaryinterauralcues.Thesearetheinteraural time(ITD),andtheinteraurallevel(ILD).Theinterauraltime istheelapsedtimebetweenthearrivalofasoundwaveatalistener'stwoears.Humans areonlysensitivetoITDsforfrequencieslowerthanabout1500Hz.The interaurallevelmeasuredindecibels(dB),istheinsoundpressure levelatalistener'stwoears.Thisiscausedmostlybytheacousticalshadowthat theheadcastsintheearthatisontheoppositesideofalistener'sheadfromasoundsource. TheILDtendstobemoreusefulathighfrequencies,wherethewavelengthissmallrelative tothelistener'sheadsize. Canalistenerlocalizeapuresinetoneinananechoicenvironmentusingonlyinteraural levelcues?Considertheazimuthalplaneforalistenerandletasourceinfrontof thelistenerzerodegreesandasourcetotherightofthelistener+90degrees. IftheILD(rightlevelminusleftlevel)risesmonotonicallyasafunctionofthisangle,thenit ispossiblethatthelistenercanuseILDcuetolocalizeasinetone.However,ifILDvs.angle isanon-monotonicfunction,thenlocalizationviatheILDwouldbeproblematicbecause 3 morethanoneazimuthallocationcouldhavethesameILD. Considerawidely-usedsphericalmodeloftheheadwitharadiusof8.75cm[15],and antipodalears.Alsoassumethatthesphericalheadishard(nosoundabsorption).Themost accessiblesphericalheadmodelwasproposedbyKuhn[27,28].Kuhngottheformulafrom Rschevkin[44],andRschevkinbuiltontheworkofMorse[38]andRayleigh[47].Thismodel solvesthewaveequationforaplanewaveundertheboundaryconditionofarigid sphere.Therefore,thesolutionexhibitsallofthepropertiesofandinterference. Figure1.1showshowtheILDbehavesasafunctionofazimuthforseveralfrequencies.The azimuthisastheincidentangleoftheincomingplanewavewheretheearsarelocated at 90 fromthefront.Forfrequenciesgreaterthanapproximately1000Hz,thecurves becometlynon-monotonic[29].Thecomplexityalsoincreaseswithfrequency,as morelocalextremaappearathigherfrequencies.Forthesakeofsimplicity,considerthe 1500-HzILDcurve.Here,thereisonlyoneextremum,whichappearsatapproximately50 withanILDofabout8dB. Whydoesthe1500Hzcurvebehavethisway?Toanswerthisquestiononeneedsto examinetheindividuallevelsattheleftandrightears.Figure1.2showsthecalculated individuallevelsoftheleftandrightearsandthebetweenthetwoasafunction ofazimuthforasoundsourceontherightside.Thelevelattherightear(theearnearthe source)slowlyandmonotonicallyincreasesbyabout3dBoverthespanof90 .Thelevel atthefar(left)eardoesnotvarymonotonically.Theleveldecreasesuntilabout50 ,where itthenbeginstoincrease.Theincreaseinlevelistheresultoftheacousticalanalogtothe opticalbrightspot,orAragospot[22].Asthesourceoftheincidentplanewaveapproaches therightsideofthehead,thesoundwavesaroundtheheadandrecombineatthe leftearincreasinglyinphase.Asaresulttheintensityattheleftearincreases. 4 Figure1.1ILDvs.azimuthforthesphericalheadmodelwithaheadradiusof8.75cm.At frequenciesgreaterthanabout1000Hz,thecurvesbecomeseriouslynon-monotonic. 5 Figure1.2Thelevelvs.azimuthfortheleftandrightearsrelativetothelevelthatwould beatthecenteroftheheadifnoheadwerepresent.TheILDisplottedvs.azimuth.They areshownforthesphericalheadmodelwithafrequencyof1500Hzandaheadradiusof 8.75cm.TheILDisshowntobenon-monotonicbecausethelevelattheleftearisnon- monotonic.Thisisbecausethelevelattheleftearincreasesafterapproximately50 ,which isduetotheacousticalbrightspot. 6 Itwasspeculatedthathumanheadswouldexhibitsimilarnon-monotonicILDcurves.If so,dolistenersrelysolelyonILDcuestolocalizethesestimuli,orcantheytakeadvantageof theindividuallevelsatthenearandfarears?Iftheycoulduseindividuallevelsatthenear andfarears,thentheywouldsuccessfullyusealloftheinformationavailabletothem,and notsimplypayattentiontotheILD.Thequestionis:Dohumanlistenerssuccessfullyuseall oftheinformationavailabletothemintheindividuallevels,ordotheysimplypayattention totheILD?Alistenerthatcantakeadvantageofindividuallevelsmaybeabletosuccessfully localizeasinetoneinananechoicroom.AlistenerthatonlyreliesonILDcueswilllikely becomeconfusedinregionswheretheslopeoftheILDcurveisnegative,andwillalsolikely beconfusedbymultipleloudspeakersthatsharesimilarILDs.Itisexpectedthatthese listenerswouldgiveincorrectand/orambiguousresponsestoidentiexperiments. Intheduplextheoryofsoundlocalization,proposedbyLordRayleigh[48],the interaurallevel(ILD)andtheinterauraltime(ITD)bothcontributeto theperceptionofthelocationofasoundsource,butcontributetlyatlowandhigh frequencies[32].Atlowfrequencies|roughlylessthan500Hz|theILDcueissmall,and theITDcuedominatesinlocalization[20,49].Attheselowerfrequencies,thewavelengthis muchlargerthanthesizeofthehumanhead,sothesoundeasilyaroundthehead, andbothearstendtobeexposedtosoundsofaverysimilarlevel.However,at1500Hz andhigher,theITDsensitivitydisappearsentirely[7,39,52],andonlytheILDcueremains. AsthewavelengthbecomesmuchsmallerthanthesizeofthehumanheadandtheIPD ˇ -limitispassed[17],itbecomesimpossibletodeterminewhichwavecyclearrivinginone earcorrespondstothesamecycleintheotherear,astherearemanycyclesinbetween.At intermediatefrequencies|roughly500to1500Hz|bothcuesplayimportantroles. 7 1.2GeneralMethods Allexperimentswereperformedinthefollowingmanner.Thelistenerwasseatedinan anechoicroom(IndustrialAcousticsCompany107840,withdimensions3mwideby4.3m longby2.4mhigh)withafrequencyoflessthan100Hz.Anarrayof13loudspeakers (RadioShackMinimus3.5,consistingofa2.5-indriverinasealedbox)wasplacedapprox- imatelyintheazimuthalplaneofthelistener'shead,whichisnedbythetwoearsand thenose.The13loudspeakers(numbered0{12)spannedtheright-frontquadrantofthe azimuthalplane,withspeakernumberzeroatanangleofzerodegreesinfrontofthelistener andspeakernumber12atanangleof90degreestotherightofthelistener.Theangular spacingbetweenthespeakerswas7.5degrees,andthegrillesofthespeakerswerelocated 111.5cmfromthelistenerformingone-forthofacircle,asshowninFig1.3.Thelistener satinachairequidistantfromtheloudspeakersandwasinstructedtofacespeakerzeroand keephisbodyandheadstillforthedurationoftheexperiment.Ametalrodlocatedjuston topofthelistener'sheadservedasareferencesothatthelistenercouldsenseheadmotion andminimizeit. ATuckerDavisSystemIIwithDD1digital-analogandanalog-digitalconverterwasused toproduceallwaveformsandwasusedtomakeallrecordings.Thesampleratewas50kHz with16bits/sample. Thestimuluswasa1500-Hzsinetoneandwaspresentedbyonlyonespeakeratatime. Eachtonehadanoveralldurationof1000ms,wherethelevelincreasedtoitsmaximumin the250msanddecreasedduringthelast250ms.Raisedcosineenvelopeswereusedfor theriseandfall.Thislongrise/falltimeisttoruleouttheuseofITD(interaural timecuesfromtheonsetofthesignal[42]. 8 Figure1.3Relationshipbetweenthelistener'sorientationandthearrayofloudspeakers. The13loudspeakers,numbered0{12,span90 .Theangularspacingbetweenloudspeakers is7 : 5 . 9 A1500-Hzsinesignalwasemployedintheexperimentbecauseifahigherfrequencywere usedinstead,thenasshowninFig.1.1,thepeakoftheILDcurvewouldmoveclosertoward 90degrees,andthereforetherewouldbelessdatapastthepeakandfeweropportunitiesto observethenon-monotonicAdditionally,theILDcurveat1500Hztendstosimply increaseandthendecrease,whereasathigherfrequenciesmoreripplesandcomplexitiestend toappearinthecurve.Ifalowerfrequencywereusedinsteadthereisapossibilitythata listenercanuseITDinthecontinuouswaveformstructureasacue.Itisgenerally believedthatfrequenciesgreaterthanabout1300Hzarehighenoughtorulethisout[52]. (Aswillbedescribedlater,theexperimentalprocedurecheckedforlocalizationusingITD cues). Fortheentiredurationofeachstimulus,recordingsweremadeinsidebothofthelistener's earcanalsusingEtymoticprobemicrophones(ER-7C).Theprobemicrophonesignalswere bymatchedandthenbyasecond(AudioBuddy) tovoltagesofafewvolts.Eachrecordingwasinthecontrollingprogramwitha verynarrowsine-cosine(matched)erynarrowbasedonone-secondrecordingsat 1500Hz|atthesamefrequencyasthestimulus.Thematchedproducestwonumbers foreachear(LandR)wherethe A 'srefertothecosineFourierintegralsandthe B 'srefer tothesineFourierintegralsinequations(1.1){(1.4). A L = 2 T Z T 0 x L ( t )cos !tdt (1.1) B L = 2 T Z T 0 x L ( t )sin !tdt (1.2) A R = 2 T Z T 0 x R ( t )cos !tdt (1.3) B R = 2 T Z T 0 x R ( t )sin !tdt (1.4) 10 Thedurationoftherecordingis T , x L ( t )and x R ( t )aretherecordingsintheleftandright ears,respectively,and ! istheangularfrequencyofthetone.The\power", P L and P R ,in theleftandrightears,respectively,werethencalculatedasshowninequations(1.5)and (1.6). P L = A 2 L + B 2 L (1.5) P R = A 2 R + B 2 R (1.6) ThentheILDissimply ILD=10log P R P L : (1.7) Thephases, ˚ L and ˚ R ,wereeachcalculatedfromthe A 'sand B 'sinequations(1.1){(1.4) as ˚ = 8 > > > > > > > > < > > > > > > > > : tan 1 B A A 0 tan 1 B A + ˇA< 0 ;B 0 tan 1 B A ˇA< 0 ;B< 0 : (1.8) ThisfunctionissimilartotheArgandatan2functionsandrestrictstheanglebetween ( ˇ;ˇ ].TheIPD,then,isjustthebetweentheleftandrightphases. IPD= ˚ R ˚ L (1.9) Forsinetones,theITDis ITD= IPD ! : (1.10) 11 Beforeeachperceptionexperiment,thelevelsoftheloudspeakerswereequalized.This wasachievedbyplacinganomni-directionalmicrophoneatthecenterofthesemi-circular arrayandmeasuringthelevelofeachspeakerfor1500-Hzsinetones.Then,compensations weremadeinthecontrollingprogram,suchthateveryspeakerproducedthesamelevelat thelistener'slocation.Overallsignallevelsforeachrunwerestandardizedbyadjustingthe gainonthepowerandaVUmeter.Theabsolutelevel(72dBA)waschosensoas tobecomfortableforthelistener,yettlyhightoperformthetasks. Therewereemalelistenersusedinthevariousexperimentstofollow.Fourofthe listenerswerebetweentheagesof20and35|labeledE,N,andX|andtheotherlistener, labeledB,was57.Alllistenershadnormalhearingthresholdswithin15dBaccordingto pure-toneaudiometry.Onelistener,M,reportedaslightasymmetricalhearinglosswiththe leftearnotassensitiveastherighteararound2000Hz.Thiswasobservedinthepuretone audiometry,howeverthelistenerwasstillwithinthe15dBrange. 1.3KEMARRecordings Asacheckontheexperimentalmethods,acousticalmeasurementsweremadeusinga KnowlesElectronicsManikinforAcousticResearch(KEMAR)[8].KEMARhasahead andtorsowithdimensionscorrespondingtothemedianhumanadult.Ithassiliconepinnae, earcanals,andacousticallyrealisticmiddleears(Zwislockicouplers).KEMARwasplaced atthecenteroftheloudspeakerarray,facingloudspeakerzero.Measurementsweremade withKEMAR'sinternalmicrophonesfor10passesthroughthe13loudspeakers(Etymotic ER11withandalsowiththeprobemicrophonesinKEMAR'searcanals. Theprobemicrophoneswereremovedandre-insertedintoKEMAR'searsfor10passes 12 throughthe13loudspeakers.Thiswasdonetosimulatevariationinpositioningthatone mightexpectinahumansubjectusingprobemicrophones.Thetenpassesthroughthe13 loudspeakersgave20recordingsperloudspeaker|aseachspeakerwasplayedtwiceinone pass.KEMAR'sILDandIPDcurveswerecalculatedforbothoftherecordingtechniques. ThecomparisonbetweenKEMAR'sinternalmicrophonesandtheprobemicrophoneswas toshowthatundertheparametersofthisexperiment,theprobemicrophonesusedwith humanlistenerswereproducingconsistentandaccuraterecordings. 1.3.1KEMARResults ItshouldbenotedthatinthereportsbelowforboththeKEMARandthehumansubject results,alllevelsandphaseshavebeensettozerowheretheincidentangleiszero.Hence, thereportthechangeintheinterauralcueswithazimuthalangle.Therecording methoddidnotmakeabsolutemeasurementsofthesoundpressurelevel.Thisisbecause theleftandrightchannelseachhadoverallgainswhichwerenotnecessarilythesameas eachother.Fortunately,theabsolutesoundpressurelevelsarenotofmuchinterest.The elevelencedependsonlyonhowthelevelschangeasincidentanglechanges, andthisiswhatisrepresentedinthefollowing Figure1.4showstheleftandrightearlevels,ILD,andIPDvs.azimuthfortheKEMAR whenusingKEMAR'sinternalmicrophonesandelectronics.TheILDandIPDgraphsare polygons,whereforeachdatapoint,themean thestandarddeviationareeachplottedand straightlinesaredrawnbetweendatapoints.Thestandarddeviationsarenearlyimpossible toseeontheseplots,indicatingexcellentreproducibility.Figure1.5showstheleftandright earlevels,ILD,andIPDvs.azimuthforKEMARwhenusingprobemicrophones.Once again,thesegraphsarepolygonsthatshowthemeans thestandarddeviations.The 13 standarddeviationsherearelargerthantheonesonFig.1.4.Thisisduetothefactthat betweentrials,theprobemicrophoneswereremovedandre-inserted(tentotalinsertions). Still,thestandarddeviationsarequitesmall,whichindicatesthatchangingthelocationof theprobetipswithintheearcanaldoesnotresultinatrencewhenmeasuring ILDandIPD. Itisnotablethatthereisverygoodagreementbetweenthetworecordingtechniquesfor theKEMAR.Thisstronglysuggeststhatusingprobemicrophonesonhumansubjectisa validwaytomeasureILD.Additionally,itisnotablethatthedatagenerallyagreewith thetrendssuggestedbythesphericalheadmodel.Bothcurvespeakataround60degrees withanILDofabout14dB.(Theadvantageofstudyingtheofvaryingmicrophone positionsintheKEMARisthatitdecouplesthemicrophone'smovementwithintheear canalsfrommovementofalistener'shead). 1.4Experiments,Results,andDiscussion 1.4.1Experiment1:IdenExperiment Inexperiment1,termedtheidenexperiment,thelistenerwasaskedtolistentothe stimuluspresentedfromaloudspeaker,andtoindicatethelocationoftheloudspeakerthat hadsoundedbycallingoutitsnumberoveranintercom.Amapoftheloudspeakerlocations wasplacedinfrontofthelistenersothatthelistenerwouldnothavetoturnhisheadto thenumbercorrespondingtosomelocationinspace.Thestimulussetconsistedof 1500-Hztonespresentedintwoone-secondintervalswith250-msraisedcosineedges.There wasa1.5-secondpausebetweenthetwointervals.Eachlistenerperformed2runs.Eachrun consistedofepassesthroughthe13loudspeakerspresentedinrandomorder,giving65 14 Figure1.4Levels,ILD,andIPDvs.azimuthforaKEMARmanikin.KEMAR'sown internalmicrophonesandelectronicswereusedfortheserecordings.Theinterauralgraphs arepolygons,wherethemean thestandarddeviationareplottedforeachloudspeaker, andstraightlinesaredrawnbetweenthepoints.However,thestandarddeviationsareso smallthatnopolygoncanbeseen.Therewere10passesthroughtheloudspeakerarray. Sinceeachloudspeakerispresentedtwiceinasinglepass,thereare20measurementsfor eachsourcenumber. 15 Figure1.5Levels,ILD,andIPDvs.azimuthforaKEMARmanikin.Probemicrophones wereinsertedintotheKEMAR'searcanalsandwereusedtomaketheserecordings.The interauralgraphsarepolygons,wherethemean thestandarddeviationsareplottedfor eachloudspeaker,andstraightlinesaredrawnbetweenthepoints.Therewere10passes throughtheloudspeakerarray.Sinceeachloudspeakerispresentedtwiceinasinglepass, thereare20measurementsforeachsourcenumber. 16 trialsperrun,and130trialsperlistener.Eachpassof13loudspeakerswascompletedbefore thenextpassbegan.Theprobemicrophoneswereusedtomakerecordingsinthelistener's ears.Becauseeachtrialpresentedtwotones,therewere20recordingsperloudspeakerfor eachlistener.Therunslastedapproximatelytentominuteseach. 1.4.1.1IdenExperimentResults InFigs.1.6{1.10,thephysicalmeasurementsappearonpanels(b),(c),and(d),whereas thelistenerresponsesappearinpanels(a),(c),and(d).Panels(c)and(d)alsodisplaythe correlationcotbetweenthephysicalmeasurementsandthelistenerresponses.For allelistenersthereisagoodcorrelationbetweentheresponsesandthemeasuredILD. However,thereisaweakercorrelationbetweentheresponsesandthemeasuredITD.For eachlistener,theresponsecurvepeaktendstooccuratthesameanglewheretheILDcurve peaks.ThisstronglysuggeststhatlistenersarelocalizingbasedononlyILD,andareeasily fooledbyspeakerstotherightofthepeakoftheirILDcurves. Panel(b)ofFigs.1.6{1.10showsthelevelsateachlistener'snearandfarears.For listenersMandX,withincreasingazimuththereisanincreaseofabout10dB,andfor listenerE,thelevelbecomesnegative.Overallthereareremarkablebetween individualsforthenearearandequallyremarkablesimilaritiesforthefarear.Thisisnota resultthatonewouldhaveanticipated.Onemighthaveexpectedthereverse.The fortheneareararetrueindividualnotmeasurementvariability. Itisparticularlyusefultoconsiderthedipineachlistener'sfarearlevel.Forthee listeners,thisdiprangesfromabout 10to 16dB.Thelocationofthedipissomewhat consistent,however.ForlistenerBandforKEMAR,thedipoccursatspeaker8(60 ),and fortheotherfourlisteners,thedipoccursatspeaker7(52 : 5 ). 17 Figure1.6ResultsfortheidenexperimentforlistenerB.Panel(a)showslistener responsesforeachtrial.Responsenumberisplottedvs.sourcenumberorazimuth.Panel (b)showsthemeanlevelforthenear(right)andfar(left)ears.Thelevelshavebeenset to0dBat0 .Errorbarsshow thestandarddeviationoverthe20measurementsateach angle.Inpanel(c),thedatapointsshowthemeanresponsenumberforeachsourcenumber, wheretheerrorbarsshow thestandarddeviationoverthe10responsesateachsource number.Theshadedpolygonrepresentsthemean standarddeviationoftheILDforeach sourcenumber.Thecorrelationcot(CC)betweenthetwosetsofdatais0.952for listenerB.Panel(d)showsthesamelistenerresponsedataaspanel(c)andalsoshows theITDmeanandstandarddeviation,whichisrepresentedbytheshadedpolygon.The correlationcot(CC)betweenthetwosetsofdatais0.512. 18 Figure1.7SameasforFig.1.6exceptforlistenerMandthecorrelationcotsfor panels(c)and(d)are0.986and0.743,respectively. 19 Figure1.8SameasforFig.1.6exceptforlistenerNandthecorrelationcotsfor panels(c)and(d)are0.945and0.293,respectively. 20 Figure1.9SameasforFig.1.6exceptforlistenerEandthecorrelationcotsfor panels(c)and(d)are0.991and0.443,respectively. 21 Figure1.10SameasforFig.1.6exceptforlistenerXandthecorrelationcotsfor panels(c)and(d)are0.920and0.740,respectively. 22 Figure1.11SpeakerResponsevs.ILD.Eachdatapointrepresentsaparticularloudspeaker foragivenlistener.Eachlistener'sspeakerresponseisplottedversusthemeanILDfora givenloudspeaker.Theerrorbarsshowplusandminusthestandarddeviationforspeaker response.ThestandarddeviationsforILDarenotshown,butaregenerallyontheorderof about1dB.Thetwocurvesaregivenbyequation1wheretheuppercurveistothese datawithaparameterof5.95,andthelowercurveistoYost's1981experimentwitha parameterof8. 23 Asshowninpanel(c)ofFigs.1.6{1.10,thepeakofthelisteners'ILDcurvesfallintotwo categories.ListenersB,MandXhadpeaksofabout11dB.ListenerNhadapeakofabout 6dB,andlistenerEhadapeakofabout7dB.Thepeakforthesphericalheadmodelwas 8dB,andforKEMARitwas14dB. ThevariationinILDforagivenlistenerandsourceisprobablyduetothemotionofthe listenersduringthetrials.Thisislikelytruebecauseoftheverysmallstandarddeviation oftheKEMAR'sILDwhenremovingandre-insertingtheprobemicrophonesover10trials. Overall,though,thehumansubject'sILDstandarddeviationswerereasonablysmall(afew decibels),ascanbeseenbytheshadedregionsinpanel(c)onFigs.1.6{1.10.Lookingat thescatterplotsinpanel(a)ofFigs.1.6{1.10,itisevidentthatthevariationinresponses islargelycausedbyvariationaroundthemean,ratherthanbyextremeoutliers. ItisinterestingthatlistenerEandespeciallylistenerNwerereluctanttogiveresponses greaterthanspeakerseven.Thismaybeduetotheofviewthatthelistenershad,as itisforlistenerstolookdirectlyatspeakersfartotherightsideoftheirhead.So althoughthesetypesoflistenersmighthavebeensuccessfulincorrectlyidentifyingsources uptoandincludingthepeakoftheirILDcurves,theystillexperiencedconfusionabout sourcesbeyondthepeak.Theselistenersmayconcedethatthebestthattheycandoisto correctlylocatethepeakoftheirILDcurve,andsimplydecidetonotrespondwithsources greaterthanaparticularspeakernumber.ListenerNseemedtodothiswhilenotminding thatallresponsesforsourcespastthepeakwouldbeincorrect.Otherlistenerstendedto stretchouttheirresponses,asiftheyassumedthatthegreatestILDthattheyheardhadto becomingfromoneofthespeakersonthefarright.Thesetstrategiesmayhavebeen theresultofvariousamountsofpracticeandknowledgegainedfromsomeofthepreliminary experiments. 24 Figure1.11displaystheresultsfromtheidenexperimentbyplottingeachlis- tener'sspeakerresponsevs.theaverageILDmeasurementforeachsourceforthatlistener. ResponsesincreaseroughlywithILD,whichistobeexpectedbasedonthecorrelationco- tsshowninpanel(c)ofFigs.1.6{1.10.However,thesedatamaybetterbedescribed byacompressionfunction.GoupellandHartmann[14]usedequation(1.11), ILD =12sgn(ILD)(1 e ILD j =p )(1.11) toYost's1981lateralizationexperiments.Thev ILD ,istheresponseforagiven ILD.Thepre-factorof12representstherangeofpossibleresponses(inthiscase0to12 loudspeakers).Theparameter, p ,wasfoundtobeequalto8forYost'sdata.Forthiscurrent experiment, p wasfoundtobe5.95.InFig.1.11,thelowercurveshows,equation(1.11) with p equalto8(forYost'sdata),andtheuppercurveshowspequalto5.95.Theupper curveappearstobeabettertotheeye.Moreimportantly,though,thedatadoappearto theshapeofthiscurve.Thisindicatesthatlistenersarebasingtheirlocalization responsesonILD. Thehighcorrelationsbetweenlisteners'responsesandILDsuggestthattheyarerelying heavilyontheILDasameansoflocalization.IstheILDtheonlypotentiallyusefulcuethat isavailabletolistenersinthisexperiment?ITDinthestructureisnotavailablebecause 1500Hzistoohigh[7].Theonsetsandaretooslowtoprovideacue.Changesin thespectrumareusedtolocalizecomplexsignals.Particularly,thepinnacausesspectral peaksandnotchesathighfrequencieswhichdisambiguatesignalsourcesthatareinfrontof, above,orbehindlisteners[6,9].Onecouldimaginethatspectralcuescouldbeusefulinthe azimuthalplaneaswell.Becauseoftheraisedcosineonsetandinthesignal,thereis 25 somespectralsplatterinthe1500Hztone.However,thereisverylittlepoweratthelowest audiblefrequencies[21],andevenlessatfrequenciesabove4kHz,wherespectralcuesare knowntobeofuse[51].Assumingthatthereisnotanynoticeablenoiseordistortionin theloudspeakers,theonlyplausiblecuesotherthentheinteraurallevelarethe individuallevelsinbothears. BecausetheILDisnon-monotonic,therearerangesofazimuthalangleswhereitis anambiguouscue.Couldalistenerwithboththeknowledgeofandtheabilitytodetect individuallevelsinbothearsbeabletodevelopastrategytodisambiguatesoundsources thatsharethesameILD?Twopossiblestrategiesutilizingtheindividuallevelsinbothears aresuggested.ThestrategybeginsbyconsideringtheILD.Foreachangleafterthe peakoftheILD,thereisacorrespondinganglebeforethepeakthatsharesthesameILD. ByconsideringILDalone,thereisnowaytodeterminewhichofthesetwoanglesisthe locationofthesoundsource.However,thisisahypotheticallistenerthatknowsthatatthis particularfrequencytheILDisambiguous.Throughexperience,thislistenerknowshow theILDbehavesvs.azimuthforanyfrequency.Furthermore,thislistenerknowshowthe nearandfarearlevelsbehavevs.azimuthforanyfrequency.Forapairofangleswithan ambiguousILD,alistenercouldconsulteithertheleftorrightlevelstoresolvetheambiguity iftheinlevelforthepairofanglesisgreaterthansome forexampleabout1dB.Eitherorbothearsmeetingthiscriteriawouldbeenoughforthis hypotheticallistenertoresolvesourcesthatareambiguousinILD.Itdoesnotmatterif theindividuallevelsareincreasingordecreasingwiththeazimuth.Thismightmakethis particularstrategylessrealisticforalistenertoemploy. Strategy2doesnotconsidertheILDatall.Insteadtheindividuallevelsaretheonly cuethatisconsidered.Becausethenon-monotonicityismostlytheresultofthechanging 26 Listener Numberofspeakers totherightofthe peak Strategy1 success rate Strategy2 success rate Correlationbetween listenerresponsesand perfectresponses B 4 25% 50% 0.49 E 5 100% 0% 0.29 M 5 60% 80% 0.64 N 5 60% 0% 0.20 X 5 80% 80% 0.76 Table1.1Independentearstrategysuccessrates,andcorrelationsbetweenlistenerre- sponsesandperfectresponses. levelinthefarear,thiscueitselfisambiguous.Alistenertryingtoresolveanambiguity inanglecausedbythenon-monotoniclevelinthefarearmayconsiderthelevelinthenear eartoresolvetheambiguity.Tomakethisstrategyeasiertoutilize,assumethatthelevel inthenearearmustincreasewithazimuth.Iftheleveldoesnotincreasewithazimuthfor apairofambiguoussources,thenthisstrategywillfail.But,thisavoidstheproblemof thelistenermemorizingalargeamountofinformationabouthowthenearearchangeswith azimuthforanyfrequency.Soassumingthatthenearearismonotonic,thenifthe inlevelbetweentheseambiguousanglesisgreaterthanthejthen thelistenershouldbeabletodeterminethecorrectangleofthesource.Thisstrategyis simplertoemploy,becausetheILDisignoredandonlythetwoindividuallevelsareanalyzed. Additionally,consultingonlytheneareartoresolvetheambiguityissimplerbecausethe neareartendstoincreasemonotonically.However,thissimplicitymeansthatstrategy2 maybelesssuccessfulthanstrategy1. Usingajust-noticableof1dB,thesuccessratesforimplementationofthese twostrategieswerecalculated.EachsourcetotherightoftheILDwaspairedwithanangle totheleftofthepeakbyinterpolatingthecue.Thentheindividuallevelsattheseangles werecompared. 27 AsshowninTable1.1,thesuccessrateusingstrategy1isquitehigh.Thisisperhaps becausethistechniqueassumesthatthelisteneristoogoodatbeingabletoanalyzethe availableinformation.Thesuccessrateforstrategy2givesmoremixedresults.Thisis probablylargelyinpartbecauseoftherequirementthatthelevelinthenearearincreases withazimuth.ForlistenerE,thenearearleveltendstodecreasewithazimuth.Thisis whythesuccessratechangedfrom100%to0%.Itininterestingthatthelistenerswiththe highestcorrelationsbetweentheirloudspeakerresponsesandperfectresponses(Listeners MandX)tendtohavethehighersuccessrateusingstrategy2.Noticethatinpanels(b) ofFigs.1.7and1.10,thefarear(squares)increasesmoredramaticallywithazimuththan forotherlisteners.Onepossibilityisthatthesetwolistenersreallyareusingthelevelsin theirnearearstoaidtheirlocalization.However,becauseofthisincreaseinnear-earlevel, theILDfortheselisteners(panelsc)decreasedtoonlyabout10dBforspeaker12,butfor theotherlistenerstheILDforspeaker12wasabout2{4dB.ThereforelistenersMandX madesmallererrorsinlocalizationatlargeazimuths.Thisalonecouldaccountfortheir betteroverallperformance.ListenerBhasamoderatesuccessrateusingstrategy2,and amoderatecorrelationbetweenhisresponsesandperfectresponse.FinallylistenersEand Nhavethelowestcorrelationsbetweentheirresponsesandperfectresponse,andalsohave thelowestsuccessrateusingstrategy2.Thisdoesnotnecessarilymeanthatlistenersare elyemployingastrategylikestrategy2,butitdoesshedsomelightonthepossibility thatlistenersareattemptingtouseindividualearlevelstolocalize,andtheonlyonesthat havesuccessinrealityaretheonesforwhichthismodelpredictssuccess.Eitherway,no listenerscouldbesaidtohavebeensuccessfullylocalizingthesestimuli;theyareclearly fooledbythenon-monotonicILD. 28 1.4.2Experiment2:DiscriminationExperiment Experiment2,termed,thediscriminationexperiment,wasidenticaltotheiden experiment,exceptthattwotloudspeakerspresentedthetwosuccessivetones.The listenerwasaskedtodetermineifthesourcemovedfromrighttoleftorfromlefttoright acrossthetwointervals.Thegoalofthediscriminationexperimentwastodiscoverwhether thesameperceptualprocessobservedintheidencationexperimentwouldalsobeobserved inthediscriminationexperiment;sp,whetherthediscriminationresultscouldbe predictedfromtheidentiresults. Thelistener'sresponsewasindicatedbypressingaleftbuttonorarightbuttonto indicatetheperceiveddirectionofmotion.Recordingsweremadeofeachstimulususing theprobemicrophones.Foreachrun,sixpairsofspeakerswerechosen,andeachpairwas presented10times.Theorderingofthe60pairswasrandom,andthedirectionthatthe sourcemovedforeachpairwasrandom.Eachlistenerperformed2runs;thereforethere were12tspeakerpairs.The12speakerpairswerenotrandomlychosen.Theywere deliberatelychosenmainlyaccordingtoexpectationsbasedontheILDcalculationsforthe sphericalheadmodel.Theintentwastochooseavarietyofspeakerpairs,witharangeof ILDchanges,andtheexpectationwasthatthelistenerwouldhavearangeofsuccessrates dependingonthespeakerpair.Successrateswereexpectedtobemostlycorrect,mostly incorrect,orambiguousresponses. Thehypothesiswasthattheresultsforthepercentageofcorrectresponsesforagiven speakerpairinthediscriminationexperimentcouldbepredictedbyusingtheresultsfrom theidenexperiment.Itwasassumedthattheunderlyingdistributionoftheinter- nalrepresentationofazimuthforeachloudspeakerwasnormalwithameanandstandard 29 deviationgivenbythemeanandstandarddeviationoftheidenexperiment.The probabilityofrespondingcorrectlythatsourceBwas|forexample|totherightofsource Awastheprobabilitythataresponselocationchosenfromtheidenndistribution forspeakerBisgreater(furthertotheright)thantheresponselocationchosenfromthe idendistributionforspeakerA.Orinotherwords,assumingthatspeakerBisto therightofspeakerA,andaparticularlistenerhasmeanresponsesforspeakersAandB, A and B ,respectively,andstandarddeviationsforspeakersAandB, ˙ A and ˙ B ,re- spectively,thenthepredictedprobability, P ,ofthelistenergivingacorrectresponseinthe discriminationexperimentis P = r 1 2 ˇ˙ 2 Z 1 0 e [ x ( B A )] 2 2 ˙ 2 dx (1.12) where ˙ 2 = ˙ 2 A + ˙ 2 B : Forexample,when( B A )islargeandpositive, P willapproach 100%.When( B A )islargeandnegative, P willapproach0%.When( B A )iszero, P is50%. 1.4.2.1DiscriminationExperimentResults Figure1.12showsthatthepredictionsofthepercentageofcorrectresponseswasreasonably successful.Ifthepredictionswereperfect,thenallofthedatawouldlieonthedashed45 line.Thiswasnotthecase,butitisimportanttorememberthateachdatapointonly representstentrials.Ifmoretrialswereperformed,itwouldnotbeunreasonabletoexpect thattherewouldbesomevariationintheactualpercentcorrectvalues.Still,asexpected, mostofthedatalieinthelower-leftandupper-rightquadrantsoftheplot.However, theredoesappeartobealargeclusteringofdataat0%and100%.Thiswouldindicate 30 Figure1.12DiscriminationExperiment.Thepredictedpercentcorrectandactualpercent correctareplottedagainsteachotherforallelisteners.Thepercentcorrectwaspredicted basedonthedatafromindividuallistenersintheidenexperiment. thatlistenersweremuchbetteratdiscriminatingthetwotonesthanwaspredictedbythe idenexperiment.Thisisunderstandable.Thediscriminationtaskseemedtobe mucheasierforlisteners.Inthediscriminationtask,alisteneronlyhastolistentostimuli fromtwosourcesatatime,andmakeajudgment,butintheidenexperiment,the listenerhastocreateamentalmapof13sourcesandtrytorememberhowthese13stimuli soundoverthecourseoftheentireidenexperiment.Thisunderstandablyleadsto alargestandarddeviationforeachspeaker'sresponsenumber.Thiswouldtendtocause 31 Figure1.13Thepercentofcorrectresponsesvs.theaverageinILDbetweenthe speakerpairs.TheinILDistakentobetherightspeaker'sILDminustheleft speaker'sILD.ThestandarddeviationsintheILDsaretypicallyontheorderof1dBor less. 32 Fig.1.12tolookasitdoes;withthepredictionsspreadsomewhatevenlyfrom0%to100%, butwithmanyoftheactualpercentagecorrectvaluesateither0%or100%. Figure1.13displaystheactualpercentcorrectvs.theaverageILDthatwasmeasured betweenthespeakerpairs.WhentheILDincreasedfromlefttoright,thelistenerstended toperformverywell.Similarly,whentheILDdecreasedfromlefttoright,thelisteners performedverypoorly.Thisiswhatwasexpected.Onlywhenthe j j waslessthan about4dB,didthelistenerstendtogiveambiguousresponses. Althoughtheidenexperimentwasnotentirelysuccessfulinpredictingthedis- criminationexperimentresults,theseresultsagreewiththeidenresultsinthat thereisnoevidencethatthelistenerswereabletorelyonanythingotherthantheILDto completethetask.Forexample,thelistenerswereunabletousetheindividualleftandright earlevelstosuccessfullyperformthediscriminationexperiment. 1.4.3Experiment3:IdenwithTrainingandFeedback Experiment Themotivationforthisexperimentwastoseeiflistenerswereabletolearntosuccessfully localizethestimuli.Fromtheresultsoftheidenexperiment,itisknownthat untrainedlistenersarenotsuccessfulatlocalization.Perhapsifthelistenersweregiven trainingandfeedbackduringtheexperiment,theirresponseswouldimprove.Intheidenti- withtrainingandfeedbackexperiment,thelistenerperformedthesametaskasin theidentionexperiment,butwaspresentedwithtrainingbeforeeachpassthroughthe 13loudspeakers,andwaspresentedwithfeedbackduringeachpass.Eachlistenerperformed eruns,eachwithtwopassesthroughthe13loudspeakersforatotaloftenpassesthrough 33 eachspeaker.Thereasontherewereonlytwopassesperrunwasthatthetrainingand feedbackcausedtherunstobelongerthanatypicalidenrun.Thetrainingbefore eachpassconsistedofplayingthe13loudspeakersinorder|fromthefronttotheright|for thelistener.Thelistenerknewthattheloudspeakerswerebeingpresentedinorderand wasinstructedtomakeanattempttousethetrainingtohisadvantage.Thenatureof thefeedbackwastoinformthelistenerastowhetherthestimuluscamefromspeakerszero throughsix,orfromspeakersseventhroughtwelve.Thefeedbackwasgivenonlyafterthe listenergavehisresponse,andwaspresentedbymeansoftwolightswhichwere locateddirectlybelowspeakerzero. 1.4.3.1IdenwithTrainingandFeedbackResults Column 1 2 3 4 5 6 x NTF1 NTF1 NTF1 NTF2 NTF2 TF1 y NTF2 TF1 TF2 TF1 TF2 TF2 Listener B 0.91 0.86 0.90 0.94 0.90 0.92 E 0.90 0.78 0.87 0.87 0.94 0.93 M 0.97 0.92 0.90 0.87 0.91 0.96 N 0.96 0.93 0.86 0.97 0.81 0.81 X 0.97 0.81 0.85 0.76 0.83 0.94 Mean 0.94 0.86 0.88 0.88 0.88 0.91 Table1.2ThePearsonproduct-momentcorrelationcotsforvarioussetsofdatafor eachlistener,aswellasthemeancotsacrossallofthelisteners.NTF1referstothe epassesoftheexperimentwithnotrainingandfeedback.NTF2referstothesecond epassesoftheexperimentwithnotrainingandfeedback.TF1standsforthee passesoftheexperimentwithtrainingandfeedback.TF2referstothesecondepasses oftheexperimentwithtrainingandfeedback.Foreachofthesehalfexperiments|which consistofepassesthroughtheloudspeakerarray|themeanresponseforeachloudspeaker wasusedinthecorrelationcalculations. AsseeninFigs.1.14{1.18,theidenionwithtrainingandfeedbackresponsesare 34 Figure1.14LoudspeakerresponsenumbersforlistenerBduringthetrainingandfeedback experiment.Therearetenresponsespersource.Thecirclesrepresentresponsesforthe epassesthroughthe13-loudspeakerarray,andthetrianglesrepresentresponsesfor thesecondepassesthroughthe13-loudspeakerarray.Thedatapointshavebeenjogged slightlyofthegridlinessothattheymayallbevisible. 35 Figure1.15SameasFig.1.14exceptforlistenerE. 36 Figure1.16SameasFig.1.14exceptforlistenerM. 37 Figure1.17SameasFig.1.14exceptforlistenerN. 38 Figure1.18SameasFig.1.14exceptforlistenerX. 39 Column 7 8 9 10 x Perfect Perfect Perfect Perfect y NTF1 NTF2 TF1 TF2 Listener B 0.35 0.60 0.56 0.37 E 0.31 0.27 0.53 0.44 M 0.58 0.70 0.64 0.77 N 0.28 0.12 0.07 0.45 X 0.80 0.72 0.79 0.75 Mean 0.46 0.48 0.52 0.56 Table1.3ThePearsonproduct-momentcorrelationcotsforvarioussetsofdatafor eachlistener,aswellasthemeancotsacrossallofthelisteners.Abbreviationsare thesameasthoseexplainedinthecaptionforTable1.2.\Perfect"referstoahypothetical listenerwhoalwaysgivesthecorrectresponsetoastimulus. non-monotonicforalllisteners,justasaretheresponseswithouttrainingandfeedback. Subjectively,thedataintheseressuggestthatlistenerswerenotsuccessfulatlocalizing thestimuliwiththeaidoftrainingandfeedback.Additionally,inlookingatthedi betweenthecircledatapointsandthetriangledatapoints,itappearsasthoughthereis notatrenceinlearningbetweentheepassesthroughthearrayand thesecondepassesthroughthearray. Tables1.2and1.3showPearsonproduct-momentcorrelationcotsforeachlistener. Thesecotswerecalculatedwiththemeanspeakerresponsesduringtheand secondhalvesofvariousexperiments.Eachcot, r x;y isthencalculatedusingequation (1.13) r x;y = 12 P i =0 ( x i x )( y i y ) s 12 P i =0 ( x i x ) 2 12 P i =0 ( y i y ) 2 (1.13) where x i and y i arethemeanresponsesforspeaker i fortheepassesindicatedinthe 40 Column 11 12 13 x Training&Feedback Training&Feedback NoTrainingandFeedback y NoTraining&Feedback Perfect Perfect Listener B 0.94 0.49 0.49 E 0.91 0.49 0.29 M 0.92 0.71 0.64 N 0.94 0.28 0.20 X 0.83 0.78 0.76 Mean 0.91 0.55 0.48 Table1.4Pearsonproduct-momentcorrelationcotsforalltenpassesthroughthe loudspeakerarray.Eachlistener'scotsareshownforvarioussetsofdataaswellas themeancotsacrossallofthelisteners. columnheadingsinTables1.2and1.3.Symbols x and y andarethemeansofthe x i and y i , respectively.InTable1.4,thecotsarecalculatedusingthemeanspeakerresponses foralltenpassesthroughtheloudspeakerarray. Overall,thereisahighcorrelationbetweentheandsecondhalvesofthetrainingand feedbackexperiment(column6),whichwouldindicatethatlistenerswerenotresponding verytlyduringthebeginningandendofthisexperiment.Thiswasalsothecase duringtheexperimentwithnotrainingandfeedback(column1).Listenerschangedtheir behaviormoreduringtrainingandfeedbackmorethanwithout,butnotbyalot(0.91versus 0.94).Thereis,however,lesscorrelationbetweenexperimentswithandwithouttraining andfeedback(columns2-5).Valuesincolumns2{5havecorrelationsthatareapproximately 0.87.Thisindicatesthatlistenersdidchangebehaviorslightlywithtrainingandfeedback, butthatthebehaviorchangelikelyhappenedearlyintheexperiment,ratherthangradually attemptingtolearnastheexperimentprogressed.Didthissmallchangeinbehaviorresult inbetterlocalization?Toanswerthisquestion,theandsecondhalvesofthetwo experimentswerealsocomparedwithahypotheticallistenerthatalwaysgivesthecorrect 41 responses.Lookingatthesecorrelations,itappearsthatthelistenersdidimproveslightly withtrainingandfeedback,butnotbyverymuch. Table1.4showsthecorrelationcotsthatwerecalculatedusingthemeanresponses acrossalltenpassesthroughtheloudspeakerarray,ratherthanthestorsecondhalfof eachexperiment.Column11comparestheexperimentwithtrainingandfeedbacktothe experimentwithnotrainingandfeedback,column12comparestheexperimentwithtraining andfeedbacktoahypotheticallistenerwhoalwaysgivescorrectresponses,andcolumn13 comparestheexperimentwithouttrainingandfeedbacktoahypotheticallistenerwhoalways givescorrectresponses.Column11'sresultsaresimilartocolumns2{5inTable2,which alsocompareexperimentswithtrainingandfeedbacktoexperimentswithouttrainingand feedback.Themeanofcolumn12(0.55)isslightlylargerthanthemeanofcolumn13(0.48), whichwouldindicatethatontheaverage,listenerswereperformingslightlybetterwith trainingandfeedbackthanwithout.ListenerEimprovedthemost(0.29to0.49),however hisperformancewasworsethanthemeanbothwithandwithouttrainingandfeedback. Althoughthecorrelationswithperfectresponsesarehigherwithtrainingandfeedbackthan without,itwouldnotbeappropriatetostatethatanylistenerwassuccessfullylocalizing duringeitheroftheseexperiments,although,themeancorrelationsincreaseintheorder thatonemightexpectthemto.Insummary,listenerswerenotabletosuccessfullylocalize thestimuliwithtrainingandfeedback. 42 1.4.4Experiment4:IdenwithAmplitudeModulation Experiment Generally,itisknownthatlistenersareunabletousetheITDinthewaveformtolocalize forfrequenciesgreaterthanabout1500Hz[7].Forthelistenersinthisexperiment,this claimissupportedbypanel(d)inFigs.6{10.Theseplotsshowpoorcorrelationbetween listeners'responsesandITD.However,itisalsoknownthatlistenersareabletolocalize onthebasisofITDintheenvelopeformodulatedhigh-frequencytones.Modulatingthe amplitudeofthesinetoneatalowfrequencyaddsenvelopetionswithpotentially usefulITDcues[23,33,46]. Theamplitudemodulationexperimentconsistedofa1500-Hztonemodulatedat100Hz. Thistonelastedfor1000mswitharise/falltimeof250ms.Thelistenercompletedtwo runswithepasseseachforatotaloftenpassesthroughthe13loudspeakers.Acoustical measurementsintheearcanalswerenotmadeduringthisexperiment. 1.4.4.1IdenwithAmplitudeModulationResults Figure1.19andTable1.5showthatsomelistenersweresuccessfulinlocalizingthestimuli, andotherswerenot.Onaverage,listenersimprovedfromdatacolumn3(0.48correlation betweenpuretoneandperfectresponses)todatacolumn2(0.86correlationbetweenampli- tudemodulationandperfectresponses).ListenerNwasthemostsuccessfullocalizerinthis experiment.ListenerN'sresponsesaremonotonic|withtheexceptionofspeaker12|and appeartolieonthe45 diagonal.Thisisinagreementwiththecorrelationbetweenampli- tudemodulationandperfectresponsesof0.99forlistenerN.Thisisadramaticimprovement overthislistener'scorrelationbetweenpuretoneandperfectresponsesof0.20.ListenerB 43 Figure1.19Eachlistener'smeanresponseforeachsourcenumberintheamplitudemodu- lationexperiment. 44 Column 1 2 3 x AmplitudeModulation AmplitudeModulation PureTone y PureTone Perfect Perfect Listener B 0.46 0.98 0.49 E 0.66 0.82 0.29 M 0.94 0.63 0.64 N 0.20 0.99 0.20 X 0.93 0.87 0.76 Mean 0.64 0.86 0.48 Table1.5Pearsonproduct-momentcorrelationcotsforvariousexperiments.The cotsarecalculatedforeachlistener,andthemeancotsareaveragedover theelisteners.Equation(1.13)isusedtocalculatethecots,exceptthatfora givenlistenerandexperiment,thedatausedinthecorrelationcalculationwerethemean responsesofthetenpassesthrougheachloudspeaker.\PureTone"referstoexperiment 1,\AmplitudeModulation"referstoexperiment4,and\Perfect"referstoahypothetical listenerwhoalwaysgivesthecorrectresponsetoastimulus. performedwellinthisexperimentalsowithacorrelationbetweenamplitudemodulationand perfectresponsesof0.98.Withtheadditionofamplitudemodulation,listenersBandN couldbesaidtobegivingmoreweighttoITDthanILD.ListenerX'sresponsesplateauat speakernumbere,ratherthanrollingoverastheydidinthepuretoneidenex- periment.ThiswouldsuggestthatlistenerXwasgivingmoderateamountsofweighttoboth ITDandILDcuesinthisexperiment.AlthoughlistenerXwasthebestpure-tonelocalizer, thislistenerstillfoundroomforimprovementwiththeadditionofamplitudemodulation. AlthoughlistenerEdidnotperformaswellasthepreviousthreelisteners(amplitudemod- ulationandperfectcorrelationof0.82),hedidimprovesubstantiallyfromthepuretone experiment(puretoneandperfectcorrelationof0.29).Thislistenerwasalsolikelygiving moderateamountsofweighttobothITDandILDcues.ListenerMwasbyfarthepoorest performer.Thiswastheonlylistenerwithacorrelationbetweenamplitudemodulationand perfectresponseswhichwasequivalenttohiscorrelationbetweenpuretonesandperfect 45 responses.Itappearsasthoughthislistenergavealmostnoweighttothenewlyintroduced ITDcues.Overall,theseresultssupporttheideathatlistenersweightITDandILDcuesin awidevarietyoftways. 1.4.5Experiment5:IdenwithNarrow-bandNoiseEx- periment Intheidenwithnarrow-bandnoiseexperiment,itispresumedthatlistenersmaybe abletouseenvelopeITDcuestolocalize,inasimilarfashiontotheamplitudemodulation experiment. Intheidenwithnarrow-bandnoiseexperiment,thesignalusedwasanarrow bandofnoisefrom1400to1600Hz(whichhasthesamebandwidthastheamplitude modulatedsignalusedinexperiment4).Thenoise x ( t ),had201componentsofequal amplitudeandrandomphaseasseeninequation(1.14),where ˚ f istherandomphasefor frequency, f . x ( t )= 1600 X f =1400 sin(2 ˇft + ˚ f )(1.14) Thewasthenscaledsothatthemagnitudeofthepeakwas+5volts,andtreated witharaisedcosinewindow,givingthealengthof1000msandthesamerise/fall timeof250msthatwasusedinthepreviousexperiments.Therewere13tnoise used|therandomphasesweretineachone|andarandomwasselected foreachstimulus.The13noisewerenormalizedsoastohaveequalpower.Asin theregularidenexperiment,thelistenercompletedatotaloftenpassesthrough the13loudspeakers.Thiswasaccomplishedintworunswithepasseseachthrough 46 Column 1 2 3 4 x Noise Noise PureTone Noise y PureTone Perfect Perfect AmplitudeModulation Listener B 0.68 0.91 0.49 0.94 E 0.56 0.86 0.29 0.94 M 0.94 0.67 0.64 0.98 N 0.20 0.99 0.20 0.99 X 0.92 0.93 0.76 0.98 Mean 0.66 0.87 0.48 0.97 Table1.6Pearsonproduct-momentcorrelationcotsforvariousexperiments.The cotsarecalculatedforeachlistener,andthemeancotsareaveragedoverthe elisteners.Equation(1.13)isusedtocalculatethecots,exceptthatforagiven listenerandexperiment,thedatausedinthecorrelationcalculationwerethemeanresponses ofthetenpassesthrougheachloudspeaker.\Noise"referstheexperimentwithnarrow-band noise,\PureTone"referstoexperiment1,\AmplitudeModulation"referstoexperiment4, and\Perfect"referstoahypotheticallistenerwhoalwaysgivesthecorrectresponsetoa stimulus. theloudspeakerarray.Acousticalmeasurementswerenotmadeinearcanalsduringthis experiment. 1.4.5.1IdenwithNarrow-bandNoiseResults Figure1.20showsthatsomelistenerswereabletolocalizesuccessfullywhileotherswerenot. Table1.6showsthattheaveragelistenerimprovedfromcolumnsix(0.48correlationbetween puretoneandperfectresponses)tocolumne(0.87correlationbetweennarrow-bandnoise andperfectresponses).Thenarrow-bandnoiseresultsaresimilartotheamplitudemod- ulationresults.Somelistenersshowedimprovementintheirabilitytosuccessfullylocalize thestimuli,whencomparedwiththeirresultsfromthe1500-Hzsine-tone.Inparticularlis- tenerNperformedquitewellinthisexperimentjusthashedidintheamplitudemodulation experiment.ListenerX'sresponsesdoappeartobemonotonic,howeveraplateaubegins 47 Figure1.20Eachlistener'smeanresponseforeachsourcenumberforthenarrow-band noiseexperiment. 48 toformaroundspeakernumbernine.ListenerEalsoshowsaplateauaroundspeakersix. Meanwhile,listenersBandMshowlittleornoimprovement.Overall,listenersperformed verysimilarlytotheamplitudemodulationexperiment(listenersBandEhadthelowest correlations,0.94,betweentheseexperiments).Generallyspeaking,itappearsthatlisteners haveawidevarietyofresultsinthisexperiment.Justasintheamplitudemodulationexper- iment,itseemsthatsomelistenersweightenvelopeITD(EITD)cuesmoreheavilythanILD cues,whileothersdotheopposite.Thedatadonotshowasystematicbetweenthe resultsfromtheamplitudemodulationexperimentandthenarrow-bandnoiseexperiment. Theaveragecorrelationbetweentheseexperimentswas0.97(fromdatacolumn7).Thisis alsosupportedbytheequivalencebetweentheaveragecorrelationsbetweenamplitudemod- ulationandperfectresponses(0.86fromdatacolumn2),andbetweennarrow-bandnoseand perfectresponses(0.87fromdatacolumn5). 1.5Theofband-widthonnon-monotonicity Sinetonesat1500Hzhavetnon-monotonicILDs.However,morerealisticlistening environmentshavebandsofnoiseofvariouswidths.Itwouldbeexpectedthatthenon- monotoniccharacteroftheILDcurveshoulddeclineasthebandwidthincreases.Forthe spherical-headmodel,Fig.1.1showsthatattfrequencies,thepeakoftheILD occursattazimuths.Additionally,theshapeofthecurve,andtheheightofthe peakchangeswithfrequency.Atlowerfrequenciestheheightofthepeakissmaller,so perhapsthesefrequenciestonotaltertheoverallILDasmuchasathigherfrequencies.The generaltrendasfrequencyincreasesis,thepeakincreases,theazimuthofthepeakincreases, andthecurveasawholeappearstobemorecomplicated.PerhapstheILDthatresults 49 fromabandcontainingfrequenciesofthisvarietywillbecomesmoothandmonotonic. TheofbandwidthonILDwasstudiedusingtheKEMARmanikin.Inaddition toa1500-Hzsinetone,thebandwidthsrangedfromaone-thirdofanoctaveto4octaves, andincreasedin1 = 3-octaveincrements.Thebandswerelogarithmicallycenteredaround 1500Hz.Thecomponentsofeachstimuluswereseparatedby1Hz.Theyhadequalampli- tudesandrandomphases.Thebandsofnoisewereplayedcontinuouslyonaloop,andwere playedinplaceofasinetoneusingthecomputerprogram. ProbemicrophoneswereusedintheKEMAR'searcanalstomaketherecordings.The ILDsforeachspeakerwerecalculatedfromthermsvoltagesintheleftandrightprobe microphones.Foreachbandofnoise,recordingswerealsomadeusingtheomni-directional microphoneinthecenteroftheloudspeakerarray.Thisallowedforesinlevel betweentloudspeakersbebecompensatedforsothatthatthelevelsintheleft andrightearscouldbecalculatedseparatelyaswell.Thelevelsatloudspeakerzerowere subtractedfromtheindividualearlevelsatallloudspeakers,andtheILDsatloudspeaker zeroweresubtractedfromtheILDsatallloudspeakers.Therefore,alllevelsandILDsare shownrelativetospeakerzero,andtheyareshownasifeachloudspeakerweretoproduce thesamelevelatthecenterofthearray.Theaboveprocedurewasdoneinplaceofthe loudspeakerequalizationprocess,whichwasusedfortheperceptualexperiments. 1.5.0.2Theofband-widthonnon-monotonicityresults Figures1.21{1.26showtheprogressioninthelevelsfortherightear,leftear,andtheILD. Themeasurementsmadeatthe1/3-octaveincrementswhicharen'tshownhererepresent onlysmallintermediatechanges.Figure1.21showsslightlytlevelsthanareshown inFigs.1.4and1.5.Forexample,thepeakILDinthesemeasurementsnowoccursat 50 Figure1.21KEMARlevelsforright,leftandILDwitha1500Hzsine-tonestimulus.The recordingsweremadeusingprobemicrophones.Levelsareshownwithrespecttospeaker zero,andareadjustedfortherelativelevelofthestimulusforeachloudspeakeratthecenter ofthearray. 51 Figure1.22KEMARlevelsforright,leftandILDwitha1/3-octave-bandstimuluscentered logarithmicallyaround1500Hz.Therecordingsweremadeusingprobemicrophones.Levels areshownwithrespecttospeakerzero,andareadjustedfortherelativelevelofthestimulus foreachloudspeakeratthecenterofthearray. 52 Figure1.23SameasFig.1.22butwitha1-octaveband. 53 Figure1.24SameasFig.1.22butwitha2-octaveband. 54 Figure1.25SameasFig.1.22butwitha3-octaveband. 55 Figure1.26SameasFig.1.22butwitha4-octaveband. 56 loudspeaker7(52 : 5 )wherethepreviousmeasurementsshowninFig.1.4showapeakILD atloudspeaker8(60 ).Additionally,theILDinFig.1.21hasamaximumvalueofabout 11dB,whereinFig.1.4themaximumvalueisabout12dB.Figure1.21alsoshowsnegative ILDsforloudspeakers2and3.NonegativeILD'soccurinFig.1.4.Theseare likelycausedbyvariationsintherepositioningoftheloudspeakerarrayandtheKEMAR intheanechoicchamber.AllmeasurementsinFigs.1.21{1.26,however,weremadewithout anyrepositioning. betweenthe1500-Hzsine-toneILDsinFig.1.21andtheone-thirdoctave bandinFig.1.22aretypically1dBorless.ThisindicatesthattheILDoftheone-third octavebandiselyequivalenttothesinetone.Figure1.23showstheone-octaveband. TheILDpeakhasmovedtospeaker8(60 ).TheILDalsonowfallsmoregraduallythan forthesine-tone.However,astrongmonotonicILDcurveisstillpresentfortheone-octave band.Thetwo-octavebandinFig.1.24continuesthistrend.Thepeakhasmovedtoa largerazimuthagain.Itisnowatspeakernine(67 : 5 ).Thispeakappearstobeaboutas broadastheone-octavepeakinFig.1.23. Thingschangeverylittleasthebandwidthisincreasedto3and4octavesinFigs. 1.25and1.26,respectively.The4-octavebandspansfrom375to6000Hz.Notonlyis theseriouslynon-monotonicILDpersistingatsuchalargebandwidth,itappearstobe stabilizingasthebandwidthincreases.Althoughthesebandsarelogarithmicallycentered around1500Hz,wherethenon-monotonicityposesthemostseriousproblemforlisteners, theseresultssuggestthatthenon-monotonicILDwillmanifestinmanysignalsbeyondsine- tonesornarrowbands.Iflistenersdoperformbetterinlocalizationasbandwidthincreases, itislikelyrelatedtotheintroductionofITDsorEITDs,ratherthanduetoachangeinthe ILD. 57 1.6MinimumAudibleAngle TheresultsfromthisexperimentmayexplainsomeresultsfromMills'1958article\Onthe MinimumAudibleAngle"[34,35].Inthisexperiment,Millsmeasuredtheminimumaudible angle(MAA)forsine-tonesinananechoicroom.Intheminimumaudibleangleexperiment alistenerheardtwotonesinsuccession.Thelistenerthenwasrequiredtoindicateifthe sourceofthesoundmovedtotheleftortheright.Foreachfrequencyandazimuth, was variedandthepercentofcorrectresponseswastabulatedforeachcondition.Mills theminimumaudibleangleasbeinghalfoftheangularseparationbetween25%correct responseand75%correctresponses.Thisisessentiallyequivalenttostatingthatalistener mustachieve75%correctresponsesataparticular ,forthisangletobeconsideredthe minimumaudibleangle. Figure1.27showsMills'minimumaudibleangleexperimentresults.Whatisinteresting isthatforhigherazimuths(60 and75 ),Millswasunabletotheminimumaudible anglefromabout1250Hzto2000Hz.Theresultsofthiscurrentstudywouldsuggestthat thisisbecausethisisjustwheretheILDvs.azimuthcurvehasanegativeslope,rather thanapositiveone.ThelistenersinthiscurrentstudyhadILDpeaksbetween55 and60 for1500Hz.ThelistenersinMills'experimentwerelikelyperceivingleft-movingsounds asright-movingsoundsandviseversa.ThiswouldexplainwhyMillswasabletoa minimumaudileangleat45 butnotat60 for1500Hz.Asthefrequencyincreasedupto approximately3000Hz,Millswasabletomeasureminimumaudibleangleagainforboth 60 and75 .Figure1.1showsforasphericalheadmodelthatasfrequencyincreases,sodoes thepeakoftheILDvs.azimuthcurve.Ifalistener'sILDpeakliesbeyond75 for3000Hz andhigher,thenitwouldbereasonabletoexpectthislistenertobeabletosuccessfully 58 Figure1.27AverageminimumaudibleanglefromMills,1958.Theesetsofdataarefor tazimuths. 59 localizethesoundsource. 1.7Conclusion Humansoundlocalizationinthehorizontalplanedependsontwointerauralcues:thein- terauraltime(ITD)andtheinteraurallevel(ILD).Forsinetoneswith frequenciesof1500Hzandhigher,humansareinsensitivetotheITD.TheILDistheonly ongoingcuethatremains.Astheazimuthalangleincreasestotherightthelevelatalis- tener'sleftear(farear)initiallydecreasesbecausetheheadcreatesanacousticalshadow. However,duetoofsoundwavesaroundthehead,thelevelinthefarearbegins toincreaseagainasthesoundsourceangleapproachestheoppositesideofthehead.Thisis knownastheacousticalbrightspot.Becauseofthis,thelevelinthefarearisnon-monotonic withazimuth.Subsequently,thelevelbetweenthetwoearsinisnon-monotonic. Thisisproblematicforlistenersbecauseanon-monotonicILDisanambiguouscue,where two,widely-separatedazimuthalanglessharethesameILDcue. Listenerswerenotabletosuccessfullylocalize1500-Hzpuretonestimuliwithazimuths greaterthanabout50degreesinaenvironment.Evenaftertrainingandfeedback, listenerswereunableuseeitherinterauralcuesorabsolutelevelinformationateachear tosuccessfullylocalize.WhenenvelopeITDcueswereintroducedtothesignal|either withamplitudemodulationorwithnarrow-bandnoise|listenersdisplayedawidevariety ofresults.Onelistenerlocalizedbothstimuliverywellandthusseemedtoheavilyweigh envelopeITDcues.Threeotherlistenersalsoshowedtlocalizationimprovement, butwerestillbyILDcuesinaddictiontotheenvelopeITDcues.Onelistener showednolocalizationimprovementandthuscontinuedtoheavilyweighILDonly.The 60 non-monotonicILDexplainsthethedatafromaminimumaudibleangleexperimentby Mills[34,35].Theminimumaudibleanglecouldnotbemeasuredatazimuthalangleswhere theILDhasanegativeslope. 61 Chapter2 AmplitudeModulationLocalization 2.1Introduction InChapter1somelistenerswereabletobfromtheintroductionofamplitudemod- ulation(AM).However,otherlistenersbtoalesserextentornotatall.Whatis responsibleforthisinperformance?Onepossibilityisthatinter-listeneranatom- icalleadtointhequalityoftheenvelopeinterauraltime (EITD)cues.Alternatively,theremaybeinter-listenerinthewaythatthein- terauralcuesareprocessed. Insoundlocalization,asoundsourceisperceivedtobeatsomelocationinspace.Insound lateralization,whichtypicallyoccurswhenlistenersareusingheadphones,asoundsource isperceivedtobesomewherealongtheinterauralaxis,orthelineconnectingalistener's ears.Listenerswearingheadphonesareabletouseenvelopeinterauraltime (EITD)tolateralizestimuli[2].Thisistrueevenathighfrequencieswhereno interauraltimecuesareavailable[3,5,23,30,33].Basedontheresultsofsinusoidally amplitude-modulated(SAM)-tonelateralizationexperimentsusingheadphones,itcanbe predictedthataddingenvelopetosignalsinfreewillhavesimilarresults. SAMtonesaresignalsthatcanbelateralizedbasedontheEITD,andarebyequation (2.1), x 0 ( t )= C [1+ m cos( ! m t + ˚ a )]sin( ! c t + ˚ c )(2.1) 62 100%amplitudemodulationspectrum Figure2.1Spectrumof100%amplitudemodulation.Thecarrierfrequency, ! c ,hasan amplitudeof C .Thesidebands,at ! c ! m and ! c + ! m ,haveamplitudesof C= 2.The phaseofthelowersidebandis ˚ c ˚ a andthephaseoftheuppersidebandis ˚ c + ˚ a . where C istheamplitudeofthecarrier, m istheamplitudemodulation(AM)percentage, ! m isthemodulationfrequency, ˚ a isthephaseoftheamplitudemodulation, ! c isthecarrier frequency,and ˚ c isthephaseofthecarrier.Inthecaseof100%amplitudemodulation, m =1.TherearethreespectralcomponentsasshowninFig.2.1. However,localizationofSAMtonesmaybeatmatterthanlateralization.Al- thoughtherehavebeenmanyheadphoneexperimentswithamplitude-modulatedstimuli, andastudyoflocalizingamplitude-modulatedhigh-frequencynoiseinfree[11],this istheknownlisteningstudyofSAM(actuallocalization).Unlikeheadphone lateralizationexperiments,inSAMlocalizationexperiments,thesignalsarenotpresented directlytothelistener'sears.Insteadtheyarepresentedfromloudspeakersourcesandthe soundwavesmustinteractwiththelistener'sanatomy.Thisresultsinamixedmodula- 63 tionsignalinthelistener'searcanals.Ingeneral,mixedmodulationincludesamplitude modulationandquasi-frequencymodulation(QFM)whichwillbeshortly. Anexperimentwasdesignedtoexaminethelocalizationof100%SAMtonesinfree Boththenatureofthephysicalcues,andtheperceptualconsequencesaretobeexamined. HowoftendolistenersinfreeencounterproblemswiththequalityoftheAM( m )? DoesthedegradationofAMqualitythestrengthoftheEITDcue?Howoftenisthe signormagnitudeoftheenvelopeITDcuemisleadingtolisteners?Thistopicofenvelope ITDlocalizationisparticularlyimportantwhenitcomestocochlearimplants[36].Forusers ofbilateralcochlearimplants,theITDcueisunavailable.Theonlytiming cueavailabletotheselistenersistheenvelopeITD. Toobtainanexpressionforasignalwithmixedmodulation,frequencymodulation (FM)isintroducedtoequation(2.1)byconsideringatime-varyingfrequencyinplaceofthe carrierfrequency, ! c ! ! ( t )= ! c + ! cos( ! m t + ˚ f )(2.2) where ! isthefrequencyexcursionand ˚ f isthephaseofthefrequencymodulation.The argumentofthesinefunctionofequation(2.1)isreplacedbythephase, Z ! ( t ) dt = Z ( ! c + ! cos( ! m t + ˚ f )) dt = ! c t + ! ! m sin( ! m t + ˚ f )+ ˚ c : (2.3) Thegeneralexpressionforamixed-modulationsignalisthen, x MM ( t )= C [1+ m cos( ! m t + ˚ a )]sin[ ! c t + sin( ! m t + ˚ f )+ ˚ c ](2.4) 64 where = ! ! m .However,foralinearsystemwithalistenerinfreethesignalsarriving intheearcanalsonlycontainspectralcomponentsfromthethreefrequenciesintheoriginal spectrum.Whenequation(2.4)isexpanded,onlytermstoorderin m and shouldbe kept.Theresultingsignalis x ( t )= C sin( ! c t + ˚ c )+ Cm cos( ! m t + ˚ a )sin( ! c t + ˚ c )+ C sin( ! m t + ˚ f )cos( ! c t + ˚ c )(2.5) whichonlyconsistsoftheoriginalspectralcomponents ! c , ! ` = ! c ! m ,and ! u = ! c + ! m where ! ` isthelowersideband,and ! u istheuppersideband.QFMwithnoAMoccurs when m =0. Todeterminethenatureofthewaveformsatalistener'searcanals,smallprobemicro- phones(EtymoticER-7C)wereused.Smallprobetubeswereinsertedintoalistener's earcanals.Becausethetubesarethin|anouter-diameterof0.95mm|thevoltagescor- respondingthedesiredsignalwereverysmallandhadanoticeableamountofunwanted electricalnoise.Becauseofthiselectricalnoise,andalsoanyunwantedacousticalnoise,it waspotentiallytodeterminethenatureofthemixed-modulationsignalsarrivingat alisteners'searcanals.Fortunately,becausethefrequenciesofthetargetSAMsignalwere known,amatcroutinecanextractonlythespectralcomponentsofinterest. Equation(2.5)canbewrittenintermsofthelower,carrier,andupperfrequenciesas showninequation(2.6). x ( t )= A c cos ! c t + B c sin ! c t + A ` cos ! ` t + B ` sin ! ` t + A u cos ! u t + B u sin ! u t (2.6) 65 Thedigitalmatcroutinecalculatesthecots, A c , B c , A ` , B ` , A u ,and B u , whichintheanalyticalrealmcanbeobtainedfromFouriertransformintegralsonthepure waveform, x ( t ),asshowninequations(2.7){(2.12). A ` = 2 T Z T 0 x ( t )cos ! ` tdt (2.7) B ` = 2 T Z T 0 x ( t )sin ! ` tdt (2.8) A c = 2 T Z T 0 x ( t )cos ! c tdt (2.9) B c = 2 T Z T 0 x ( t )sin ! c tdt (2.10) A u = 2 T Z T 0 x ( t )cos ! u tdt (2.11) B u = 2 T Z T 0 x ( t )sin ! u tdt (2.12) Therelationshipsbetweenthesecotsand C , m , , ˚ c , ˚ a ,and ˚ f areasfollowsin equations(2.13){(2.24). C = q A 2 c + B 2 c (2.13) tan ˚ c = A c B c (2.14) tan ˚ a = ( A u A ` )cos ˚ c ( B u B ` )sin ˚ c ( B u + B ` )cos ˚ c +( A u + A ` )sin ˚ c (2.15) tan ˚ f = ( A u + A ` )cos ˚ c ( B u + B ` )sin ˚ c ( B u B ` )cos ˚ c +( A u A ` )sin ˚ c (2.16) Thesolutionsfor m and areasfollows. m = 2 C B u sin( ˚ c + ˚ f ) A u cos( ˚ c + ˚ f ) sin( ˚ f ˚ a ) (2.17) 66 = 2 C A u cos( ˚ c + ˚ a ) B u sin( ˚ c + ˚ a ) sin( ˚ f ˚ a ) (2.18) Inthecasewheresin( ˚ f ˚ a )=0, m and become m = A u C sin( ˚ c + ˚ a ) + A ` C sin( ˚ c ˚ a ) (2.19) and = A u C sin( ˚ c + ˚ a ) + A ` C sin( ˚ c ˚ f ) : (2.20) Whenbothsin( ˚ f ˚ a )=0andsin( ˚ c + ˚ a )=0,thesolutionfor m is m = B u C cos( ˚ c + ˚ a ) B ` C cos( ˚ c ˚ a ) : (2.21) Whenbothsin( ˚ f ˚ a )=0andsin( ˚ c ˚ a )=0,thesolutionfor m is m = B u C cos( ˚ c ˚ a ) + B ` C cos( ˚ c + ˚ a ) : (2.22) Whenbothsin( ˚ f ˚ a )=0andsin( ˚ c + ˚ f )=0,thesolutionfor is = B u C cos( ˚ c + ˚ f ) B ` C cos( ˚ c ˚ f ) : (2.23) Whenbothsin( ˚ f ˚ a )=0andsin( ˚ c ˚ f )=0,thesolutionfor is = B u C cos( ˚ c ˚ f ) B ` C cos( ˚ c + ˚ f ) : (2.24) Itistheinterauraltimeintheenvelopeofthesignalsthatisofperceptual 67 importanceathighfrequencies[23].Theenvelope, E ( t ),canbeobtainedbycalculatingthe Hilberttransformofthesignal,^ x ,andaddingittotheoriginalsignalinquadratureasshown inequation(2.25)[16]. E 2 ( t )= x 2 ( t )+^ x 2 ( t )(2.25) Then,fromthemixed-modulationsignalfromequation(2.6),theenvelopeisgivenby E 2 ( t )=[ A c +( A u + A ` )cos ! m t +( B u B ` )sin ! m t ] 2 +[ B c +( B u + B ` )cos ! m t +( A ` A u )sin ! m t ] 2 : (2.26) 2.1.1Problem1:AMQuality Figure2.2showsanSAMtonewith100%AM, m =1.However,duetoanatomical infreealistenerwillnotgenerallyheara100%SAMtonewith m =1.Therewill generallybesomeQFMpresentinthesignal.Alistenershouldexperienceover-modulation ( m> 1)orunder-modulation( m< 1).Towhatextentthereisover-orunder-modulation willdependonanindividual'sanatomy,thesoundsourceangle,thefrequency,andthe modulationfrequency.Thedepthofmodulation|particularlyifthereisunder-modulation| mayhaveanonalistener'sabilitytoutilizetheEITDcuetolocalizethesourceof sound.Furthermore,thereisnoreasontoexpectthatthemixedmodulationparameters, including m ,willbethesameinalistener'sleftandrightears.Inthissituation,theleft andrightsignalsmaybecometoodissimilarfortheenvelopeITDstobeofuse[37]. ForapureSAMtonefrom(2.1),thedepthofmodulationisdeterminedonlyby m . However,inmixedmodulationtheQFM, ,canalsocttheoveralldepthofmodulation andmaybeperceptuallyimportant.Soitisusefultoconsideranothermeasureofthedepth ofmodulation.Theenvelopemodulationfraction, m e inequation(2.27),isthesameas 68 2000Hzcarrier,100HzAM Figure2.2SAMsignalwith100%AM.Thecarrierfrequencyis2000Hzandthemodulation frequencyis100Hz.NoQFMispresentinthissignal.Inthisexample,thesignalsinthe twoearsareidentical. 69 m forapureSAMtoneofarbitrarymodulationdepth.Itistheratioofthe betweenthemaximumandminimumvaluesoftheenvelopeandtwicetheamplitudeofthe carrierfrequency.Theenvelopeisnevernegative,butcanbezeroforcasesof100%AM orover-modulation.Over-modulationresultsinasecondarypeakintheenvelopebetween theprimarypeaks.Althoughthispeakmaybeperceptuallyrelevant,itisnotexplicitly includedinthecalculationofequation(2.27). m e = max( E ) min( E ) 2 C (2.27) ThedegradationofAMqualitycouldpotentiallyposeaproblemtolistenersfortworea- sons.First,ifthesignalisunder-modulated,theEITDcuewilllikelybeweakerandbecome non-existentasthemodulationapproacheszero.Second,theshapeoftheenvelopemaynot bethesameinalistener'sleftandrightears.Howmightthebinauralsystemprocessand comparethetimerencebetweentwoenvelopesiftheyhavetshapes?Figure2.3 showsanexampleofthewaveformfora2000-Hzcarriersignal.Inthisinterval,thesignal inthelistener'sleftearisunder-modulated,whilethesignalinthelistener'srightearis overmodulated.Figure2.4showsanotherexampleofawaveformwitha3000-Hzcarrier signal.Inthisinterval,anasymmetricalmodulationpatternisseeninbothears.Theleft earsignalcontainsalargeamountofQFM,with =0 : 56.Thesemixedmodulationwave- formexamples|whichsubstantiallyfromSAMsignalslistenersheardinheadphone experiments|arenotnecessarilyrepresentativeofarandomlyselectedrecordingfromthis experiment,butdoillustratethesomeofthemoreunusualbinauralsignalsalistenerwill encounterwhenlisteningtoSAMtonesinfree 70 2000Hzcarrier,100HzAM Figure2.3Leftandrightwaveformsofarbitraryunitscalculatedfromearcanalrecordings. Thecarrierfrequencywas2000Hzandthemodulationfrequencywas100Hz.Valuesfor theamplitudemodulation,QFM,andenvelopemodulationfractionareshownontheright by m , ,and m e ,respectively. 71 3000Hzcarrier,100HzAM Figure2.4Leftandrightwaveformsofarbitraryunitscalculatedfromearcanalrecordings.. Thecarrierfrequencywas3000Hzandthemodulationfrequencywas100Hz.Valuesfor theamplitudemodulation,QFM,andenvelopemodulationfractionareshownontheright by m , ,and m e ,respectively. 72 2.1.2Problem2:EITDGroupDelay TheeITDisthephasedelayoftheinterauralphaseshownbythe simplerelationshipinequation(2.28). ITD= 1 ! IPD(2.28) TheITDandIPDrelationshipislinearandmonotonic,sothecueisalways consistentinsignandmagnitudewiththeIPDandITD.TheenvelopeITDcue,however, isthegroupdelayoftheIPD[26]asshowninequation(2.29) EITD= d d! IPD(2.29) TheenvelopeITDcuedependsontheslopeoftheIPDwithrespecttofrequency.There isnoreasontoexpectapriorithatthisfunctionwillbelinearormonotonicforahuman headinfreeTherefore,theEITDcuemaysometimesbeincorrectinsignforagiven azimuth.ThiswouldalsomeanthattheEITDcuemaynotincreasemonotonicallywith azimuthandmaymisleadlisteners. 2.1.2.1Problem2:KEMARGroupDelay MeasurementsweremadeinananechoicroomwiththeKEMARmanikindescribedin Chapter1.Figure2.5showsthewrappedIPDsmeasuredoverafrequencyrangeof1800{ 2800Hzwithasoundsourceatanazimuthof+60degrees.Onthewhole,theIPDincreases withfrequency,buttheincreaseisnon-monotonic.Therearerangeswheretheslopeisvery steep,suchaslineA,andrangeswheretheslopeisnegative,suchaslineB.TheEITDof 73 GroupDelay Figure2.5WrappedIPD( 180 to+180 )vs.carrierfrequency.Measurementsweremade intheanechoicroomwiththe1-marray.Theazimuthwas60 .Thesignalsweresinetones at10Hzintervals. 74 PositiveGroupDelay Figure2.6Left(blue)andright(red)waveformsandenvelopesofKEMARrecordingsof SAMtonesinfreeThesignalwasa2240Hzcarrierwithamodulationfrequencyof 40Hz. anSAMtoneisthegroupdelay,ortheslopeoftheIPD. ToinvestigateregionsAandB,SAMtoneswithamodulationfrequency, f m =40Hz, werepresentedfromanazimuthof60 .Figure2.6showstheleftandrightwaveformsforthe SAMtonewithacarrierfrequencyof f c =2240Hz,correspondingtolineA.Theenvelope oftherightear(red)isleadingtheenvelopeoftheleftear(blue),asisexpected.Figure2.7, showstheleftandrightwaveformsfortheSAMtonewithacarrierfrequencyof f c =2325Hz, correspondingtolineB.Here,theenvelopeoftheleftear(blue)isleadingtheenvelopeof therightear(red),eventhoughthethesourceisontheright.Thesemeasurementsshow thatforSAMtones,theEITDcanbeamisleadingcueinfree 75 NegativeGroupDelay Figure2.7Left(blue)andright(red)waveformsandenvelopesofKEMARrecordingsof SAMtonesinfreeThesignalwasa2325Hzcarrierwithamodulationfrequencyof 40Hz. 76 2.2Methods 2.2.1ExperimentalSetup TheexperimentalsetupwasthesameasinChapter1exceptforthefollowing.Thecon- structionofthearraywast.Thegrillsofthespeakerswerelocated197cmfromthe listener,butotherwisehadthesameangularspacingasshowninFig.1.3.Forthenewarray, theloudspeakersweremountedonfourpiecesofwoodenoringwithvelcro.Thepieces ofwoodenoringweresupportedbymicrophonestands.Thepreviousarrayheightwas slightlyabovetheazimuthalplaneforatypicallistener,butthenewarraywasloweredso thattheloudspeakerconeswereintheazimuthalplaneforthelisteners.Thisnewarraywas designedtoreducescatteringofthemountinghardware.Additionally,becausethearray waslarger,thereshouldhavebeenlessscatteringofneighboringloudspeakers.Finally, theincreaseddistancebetweenthesourcesandthelistenerresultedinabetterrealization ofthelimitthantheold111.5-cmarray[41,45]. Directlybehindthelistenerwasamaskingloudspeaker(BostonAcousticsA40).The loudspeakerfaceduptowardthelistener'shead.Amaskersignal,describedbelow,was producedbyaCDplayer. 2.2.2ExperimentalConditionsandProcedure Thereweresixexperimentalconditionsforeachlistener,bytsignalsthat werepresented.Theseconsistedofsinetonesat2kHz,3kHz,and4kHz,andsinusoidally amplitudemodulatedtoneswiththesamecarrierfrequenciesandamodulationrateof100Hz (100%modulation).Themodulationfrequencyof100Hzisneartheregionaround128Hz showntobemostsensitive[4,5,10].Acriticalbandisarangeoffrequenciesthatissmall 77 enoughthatauditoryprocessingisnotindependentfortfrequencies.The100-Hz modulationfrequencywassmallenoughtoensurethatallofthespectralcomponentswere inthesamecriticalbandforeachofthethreecarrierfrequencies[13,24,25].Thereforethe SAMtoneswereperceivedassingleacousticalentities.Thestimulihadatotaldurationof 1000ms.Therewasa250-mslinearrise/fallatthebeginningsandendsonthesignals.The targetstimulihadanaveragelevelof65dBAasmeasuredatthelocationofthelistener.In eachtrialthelevelrovedrandomlyby+2,+1,0, 1,or 2dB.Thiswasdonetoensurethat thelistenercouldnotuseanylevelbetweenindividualloudspeakersornon-linear distortionproductstoidentifythesource. Themaskersignalwasintendedtomaskdistortionproductsat100and 200Hz,whichmightbeanalternativelocalizationcue.Itconsistedoffrequencieswithan evennumberofhertzfrom50to250Hz.Thecomponentshadequalamplitudeandrandom phase.This0.5-speriodsignalwasplayedcontinuouslythroughoutthecourseofarunwith alevelof50dBCasmeasuredatthelocationofthelistener'shead.Tomeasurethepower ofthemaskertoneinthelisteners'earcanals,eachrunbeganwithasilenttrialwhere recordingsweremadewhileonlythemaskerwason. Atthebeginningofeachrun,acalibrationsinetonefromspeakerzerowasturnedonwhile theexperimenterviewedtheprobemicrophonesignalsonanoscilloscope.Theexperimenter instructedthelistenertoadjusthisorherheadtoensurethatthe-structureIPDwasas closetozeroaspossible.Thiswasdoneundertheconstraintthatthelistenerfeltcont thatheorshewasfacingspeakerzero.Additionally,theexperimenteradjustedthegainson thetoensurethattheILDwaswithin1dBofzero. Arunconsistedof5randompassesthroughthe13-loudspeakerarray.Ineachtrial,there were2intervalspresentedbythesameloudspeaker.Therewasapauseofabout1second 78 betweenthe2intervals.Thelistenerwasthenaskedtoverballyrespondwithaloudspeaker number.ThediagraminFig.1.3waslocatedjustbelowspeakerzerosothatthelistener couldjudgespeakernumberswithoutturninghisorherhead.Listenerswereallowedto respondwithnegativespeakernumbersifthesourcewasperceivedtobeontheirleft,and withsourcenumbersgreaterthan12(orlessthan 12)ifthesourcewasperceivedtobe behindthem. Eachlistenercompleted2runsforeachstimuli,whichresultedin10trialsand20binaural recordingsforeachstimuli/loudspeakercombination. 2.2.3Listeners Therewere5listeners.ListenerBwasamaleaged59.ListenersC,M,andLweremales aged20{25.ListenerVwasafemaleaged19.ListenerBwastheonlylistenerthatalso participatedintheexperimentsfromChapter1.ListenersM,L,andVhadnormalhearing thresholdswithin15dBofaudiometriczerooutto8kHz.ListenerBhadamildhearing losstypicalofmaleshisage,butnormalthresholdsatthefrequenciesoftheseexperiments. ListenerChadnormalhearingthresholdsexceptforabout20dBofhearinglossinhisleft earbetween1.5and4kHz. 2.2.4ComputerAnalysisofSignals Theanalysisoftherecordingswaslimitedtothehalfsecondfrom256msto756ms,andat asamplerateof50kHz,contained25,000samplesperchannel.Thiseliminatedthe250-ms rise/falltimesatthebeginningsandendsofthesignals,andaccountedthe6-mslagtime forthesoundtotravelfromtheloudspeakerstothelistener. 79 Therawrecordings, x raw ,containedelectricalnoisefromtheacoustical noise|includingthecontinuousnoiseofthemasker|anddistortion.Toeliminatethe ofthenoiseanddistortion,thesignalsweredigitallyBecausethemaskerhada periodof500ms,itswastotallynegatedbythe500-msmatcUsingthe discrete-timeequivalentsofequations(2.7){(2.12),sixmatceringcotswere obtainedforbothearsforeachrawrecording, x raw .Thesecotswerethenusedto calculatethemodelwaveform, x model ,usingequation(2.6).Themodelwaveformpower, P model ,wascalculatedforeachearwithequation(2.30). P model = A 2 ` + B 2 ` + A 2 c + B 2 c + A 2 u + B 2 u (2.30) Thepowerofthenoiseplusdistortionintherawrecordingswasas P N + D ,raw = 1 25000 25000 X t =1 ( x raw [ t ] x model [ t ]) 2 : (2.31) Sincetherawrecordingsalsocontainednoisefromthemaskerandsomeelec- tricalnoise,thepowerofthemaskerrecordings, P masker ,wassubtractedfromthepowerof thenoiseplusdistortionintherawrecordings.Thisleftthenoiseanddistortionpoweronly duetoloudspeakerdistortionorothersourcesofsoundintheroom, P N + D . P N + D = 1 25000 25000 X t =1 ( x raw [ t ] x model [ t ]) 2 P masker (2.32) Finally,thepercentageofnoiseplusdistortionwascalculatedas 100( N + D )= P N + D P model : (2.33) 80 Recordingsandresponsesfortrialswherethenoiseplusdistortionwasgreaterthanorequal to10%werediscardedfromfurtheranalysis. Inadditiontothemodelwaveformandthemodelenvelope, C , m , , ˚ c , ˚ a ,and ˚ f werecalculatedusingthecotsandequations(2.13){(2.24)and m e using(2.27). ThemodelILDswerecalculatedusingmodelpowersfromequation(2.30). ILD=10log P model,right P model,left (2.34) ThevalueoftheILDforzerodegreesazimuthwassubtractedfromtheILDforeveryazimuth. TheILDsasrecordedhadasmallbecauseoftheunknownrenceingainbetween theleftandrightchannels. Themodelenvelopeswerecalculatedusingthematccotsandequation (2.26).EnvelopeITDswerecalculatedusingacrosscorrelation, ( ˝ ),asafunctionoflag time, ˝ ,betweentheleftandrightmodelenvelopes, E ` and E r ,respectively. [ ˝ ]= 8 > > > > > > > > > > > > > > > > > > > < > > > > > > > > > > > > > > > > > > > : 25000 ˝ P t =1 E ` [ t + ˝ ] E r [ t ] s 25000 P t =1 E 2 ` [ t ] 25000 P t =1 E 2 r [ t ] ˝ 0 25000+ ˝ P t =1 E ` [ t ] E r [ t + ˝ ] s 25000 P t =1 E 2 ` [ t ] 25000 P t =1 E 2 r [ t ] ˝< 0 (2.35) Giventhatthemodulationfrequencyis f m = ! m = 2 ˇ =100Hz,andthatthecarrierfrequen- ciesaremultiplesof100Hz,theperiodofthesignalsandenvelopesshouldbe T =1 =f m . 81 Neglectingthesmallend-ofthecross-correlationcalculation, [ ˝ ]mustalso haveaperiodof T .Therefore,themaximumof [ ˝ ]wassearchedforoverarangeof T= 2 ˝ T= 2.Thevalueof [ ˝ ]atthepeakistheenvelopecoherence,andthecorre- spondingindexedtime, ˝ ,istheenvelopeITD. 2.3ResultsandAnalysis Asmallnumberoftrialswereremovedfromtheanalysisduetonoiseplusdistortionper- centagesthatweregreaterthanorequalto10%ineitherearofeitherintervalofthetrial.In all,9trialsoutofagrandtotalof3,900wereremoved.Themeansandstandarddeviations oftheresponseswerecalculatedforeachloudspeakeracrossthetworunsforaparticular listener,frequency,andwaveform(usually10responses).Similarly,themeansandstandard deviationsofallphysicalquantitiesfromtheearcanalrecordingswerecalculatedforeach loudspeakeracrossthetworunsofaparticularlistener,frequency,andwaveform(usually 20intervals).Incaseswherealistenerrespondedwithasourcelocationfrombehind,the recordedresponseswereedtothefrontsymmetricallyacrossspeaker12at90 azimuth. 2.3.1PhysicalResults 2.3.1.1Problem1:AMQualityResults Theamplitudemodulation, m ,variedsomewhatdependingonlistenerandfrequency.Most ofthevaluesremainedclosetounity.Theleftear,whichwasontheoppositesideofthe loudspeakerarray,experiencedmorevariationthantherightear.Thevariationalsotended tobegreatestwhentheleftearwasintheacousticaldarkspot.Thiswouldcorrespond totheazimuthwherethelevelintheleftearwasthesmallest,andcorrespondswellto 82 thelocationofthepeakoftheILD.Nearthisazimuth,theintensitychangesgreatlywith azimuth,soitislikelythatthethreefrequencycomponentsoftheSAMtoneexperienced thegreatestamountofdisparityinhereaswell.Therefore m deviatedfrom1the mostinthesecases. Figure2.8isahistogramoftheamplitudemodulation, m ,forallloudspeakers,listeners, andfrequencies.Althoughthemodalvaluefor m isnear1forthesedistributions,there arelargenumberofcaseswithanotabledeviationfrom1,althoughnevermuchlowerthan about0.75orhigherthan1.75.Theleftearexhibitsmoreover-modulation,includingone outliernear3.Thebulkoftheleft(far)eardistributioniswiderthantheright(near).This islikelybecausethemagnitudesandphasesofthethreefrequencycomponentsbecomemore scrambledwhenaroundthelistener'shead. Table2.1indicateshowfrequentlyover-modulationoccurredineachearforthethree carrierfrequenciesandacrossallfrequencies.Bothearsexperiencedsomedegreeover- modulationthemajorityofthetime.Atlowercarrierfrequencies,theover-modulation occurredmorefrequently.Thismaybeunderstoodduetothefactthatthe100-Hzspacing betweenthefrequencycomponentswasasmallerratioofthecarrierfrequencyathigherfre- quencies.Therefore,assumingconstant-Qthethreecomponentswouldhavebeen inamoresimilarmannerathighercarrierfrequency,andtheoriginalwaveform wouldhavebeenbetterpreserved. Figure2.9isahistogramoftheQFM, ,foralllistenersandallfrequencies.The histogramsforbothearspeakaround0 : 1to0 : 2,andtherearenotmanyinstanceswhere exceeds0 : 4.However,theQFMintheleft(far)earfrequentlydeviatesdrasticallyfrom zero|thevalueof inthesignalplayedbytheloudspeakers.Giventhattheleftearis onthefarsideoftheheadforeachloudspeakerexceptthefrontloudspeaker,thisisnot 83 m meansacrossintervalsforallloudspeakers,listeners,andfrequencies Figure2.8Amplitudemodulation, m ,meansforeachloudspeaker,frequency,andlistener. Histogramsfortheleftandrightearsaredisplayedseparately.Eachhistogramcontainsa totalof195values. Percentof m means > 100% carrierfrequency leftear rightear 2kHz 71% 86% 3kHz 69% 66% 4kHz 23% 38% all 57% 60% Table2.1Thepercentageofamplitudemodulation, m ,means > 100%.Themeansare calculatedacrossidenticalintervalconditions.Thedataarecombinedacrossalllisteners andazimuths. 84 meansacrossintervalsforallloudspeakers,listeners,andfrequencies Figure2.9QFM, ,meansforeachloudspeaker,frequency,andlistener.Histogramsfor theleftandrightearsaredisplayedseparately.Eachhistogramcontainsatotalof195values. surprising.Althoughthectsoftazimuths,carrierfrequencies,andlistenersare notshowninthishistogram,what'sapparentisthatinSAMlistening,somedegree ofQFMisalwaysproducedintheearcanals. 2.3.1.2Problem2:EnvelopeITDResults ThedistributionofenvelopeITDsinthisexperimentisshowninFig.2.10.Thesignofthe EITDissubstantiallypositivewithamodalvalueofabout600 ,whichisconsistentwith 85 EnvelopeITDmeansacrossintervalsforallloudspeakers,listeners,andfrequencies Figure2.10HistogramofEnvelopeITDmeansforeachloudspeaker,frequencyandlistener. thesourceazimuths.Thereare,however,anumberofnegativeEITDs,whichoccurred14% ofthetime.Thesecueshavethewrongsignandshouldthereforebemisleadingcuesto listeners. 2.3.1.3KEMARHeadphonesMeasurements Sincetherehasbeenagreatdealofheadphone-basedAMresearch,itisusefultoknowto whatextenttheproblemofAMqualityexistswhenAMtonesarepresentedoverheadphones. WhilethedegradationofAMqualityinfreeappearstobeattributableto 86 KEMARSAMHeadphoneRecordings carrierfrequency ear m m e 2kHz left 1.0156 0.1629 1.0057 right 1.0072 0.1566 1.0012 3kHz left 0.9998 0.0925 0.9974 right 1.0130 0.1071 1.0053 4kHz left 0.9851 0.0627 0.9762 right 0.9939 0.0439 0.9882 Table2.2Valuesof m , ,and m e calculatedfromrecordingsofKEMARmanikin(large pinna)wearingheadphones(SennheiserHD535).Thesignalswere100%SAMtoneswith carrierfrequenciesof2,3,and4kHz,andamodulationfrequencyof100Hz.Themonaural parameterswerecalculatedusingmatched aroundthehead,itmaybethatsomedegradationcouldalsooccurwhilelisteningover headphones.Forover-earheadphones,thiswouldlikelybeduetofromthepinna.In ordertodeterminethemagnitudeofthiserecordingsweremadeofSAMtoneswiththe KEMARmanikin'sownears(largepinna)whilewearingover-earheadphones(Sennheiser HD535).The2,3,and4kHzSAMtoneswith100HzAMwerepresentedtobothears. TheresultsoftheKEMARheadphonesmeasurementsareshowninTable2.2.Allofthe valuesof m arewithin2%ofthesignalvalueof1 : 0000.Similarly,allofthevaluesof m e arewithin3%ofthesignalvalueof1 : 0000.Bycontrast,theQFM, arequiteabitlarger thanthesignalvalueof0.Thedeviationfrom0appearstodecreaseasthecarrierfrequency increases.Thiscouldbeunderstoodbyconsideringthatasthemodulationfrequencyremains constant,thesidebandsmoveclosertothecarrierfrequencyonalogarithmicscale.Therefore themagnitudesandphasesofthefrequencycomponentsarechangedlessinrelationtoeach otherathigherfrequencies.WhiletheQFMvaluesmeasuredfromtheKEMARwearing headphonesarelargerthanexpected,theyarenotnearlyaslargeassomeofthevalues foundinldforhumanlisteners(Fig.2.9). 87 2.3.2PerceptualResults 2.3.2.1InterauralCuesandResponses ListenerresponseswhenlocalizingtheSAMtonesandprobemicrophonemeasurementsmade atthesametimeareshowninFigs.2.11{2.25.Thenon-monotonicILD[31]isobservablein thesineandAMtrialsforeachlistenerandfrequency.Aspredictedbythesphericalhead model(Chapter1),asthefrequencyincreasestheILDtendstopeakatlargerazimuths, andthepeaktendstogrowaswell.Inthesinetonetrials|(a)plots|listenersareclearly misleadbythiscue.ThisisespeciallyevidentatlargesourceangleswheretheILDissmall. TheresponsesforthesinetonearehighlycorrelatedwiththeILDforalllistenersandall frequenciesasseeninthe(a)plots.ThisiseasilyexplainedbythefactthattheILDisthe onlyavailableinterauralcueavailabletolistenersforthesetrials. TheILDsintheAMtrials|(b)plots|aresimilartothesine-tonetrials,butdi somewhat.TherearetwopossiblesourcesoftheseThestisthatthelistener orientationcanchangeslightlybetweentrials.Itshouldalsobenotedthatthevariabilityin ILDwithinplots(a)and(b)maybeattributabletochangesinlistenerorientationbetween thetwotrialsforagivencondition.Additionally,alistenermayhavemadesmallmovements duringasinglerun.ThesecondpossiblesourceforILDencesbetweensineandAM runsisthatthesignalsaret.AlthoughtheSAMtone'scarrierfrequencyisthesame asthesinetone,thetwosidebandsat 100Hzwillbehaveslightlytly.Despitethe smallinILD,onecanseethatthetrendisthatresponsesdonotcorrelateaswell withtheAMILDastheydowiththesineILD.Thecorrelationisstillstrong,however.A summaryofthecorrelationsbetweenresponseandILDforAMandsinetonesisshownin Fig.2.30. 88 ThelowercorrelationsintheAMtrialsmaybecausedbytheavailabilityofanalternative ITDcue|sptheenvelopeITD(EITD).The(c)plotsshowtheEITDalongwith thesameresponsesfromthe(b)plots.LiketheILD,theEITDisnon-monotonic.TheEITD isnegativeinsomecases,althoughneverwithaverylargemagnitude.Thiswasaddressed above(Problem2:thegroupdelayoftheIPDbeingnegativeinsomecases).Thistendsto occuratsmallerazimuthalangles,andthetrendoftheEITDistorisewithazimuth. Atsmallerazimuths,theEITDisusuallynotverylarge.AtthelargerazimuthstheEITD tendstoreachlargerpositivevalues.LargerazimuthsarealsowheretheILDhasdecreased tly,whereastheEITDremainsrelativelylarge.Whencomparingresponsesbetween thesinetonetrialsandtheAMtrials,onecanseethatthebiggestchangestendtooccurin thesesituations.DespitetherelativelylowcorrelationsbetweenAMresponsesandEITD, theadditionofAMhasinsomecasesimprovedlisteners'mostseriouserrors.TheEITD happenstobethemostusefulexactlywheretheILDcuefails.Therearetwoclearexamples ofthis.InFig.2.11,at2kHzlistenerBshowsimprovementforsourcenumbers11and12 wheretheILDdecreasessharplyandtheEITDisatitslargest.Themeanresponsesfor theseloudspeakersincreasebyabout30degrees.Similarly,inFig.2.16at4kHzlistenerC showsimprovementforsourcenumbers10,11,and12. 89 ListenerB,2kHz (a)(b) (c) Figure2.11Responsesandprobe-microphonemeasurementsforlistenerB,2kHz.The sourcenumbersspantherightfrontquadrant.ThePearsonproduct-moment(PPM)cor- relationcotisshownbetweenthemeanresponsesandtheinterauralcues.(a)Sine tone:Circlesindicatemeanresponsesanderrorbarsaretwostandarddeviationsinoverall length.ThehatchedregionshowstheILD.Itiscenteredonthemeanandistwostandard deviationshigh.(b)SAMtone:Circlesindicatemeanresponsesanderrorbarsaretwostan- darddeviationsinoveralllength.ThehatchedregionshowstheILD.Itiscenteredonthe meanandistwostandarddeviationshigh.(c)SAMtone:Circlesindicatemeanresponses anderrorbarsaretwostandarddeviationsinoveralllength.Thehatchedregionshowsthe EITD.Itiscenteredonthemeanandistwostandarddeviationshigh. 90 ListenerB,3kHz (a)(b) (c) Figure2.12SameasFig.2.11butforlistenerBat3kHz. 91 ListenerB,4kHz (a)(b) (c) Figure2.13SameasFig.2.11butforlistenerBat4kHz. 92 ListenerC,2kHz (a)(b) (c) Figure2.14SameasFig.2.11butforlistenerCat2kHz. 93 ListenerC,3kHz (a)(b) (c) Figure2.15SameasFig.2.11butforlistenerCat3kHz. 94 ListenerC,4kHz (a)(b) (c) Figure2.16SameasFig.2.11butforlistenerCat4kHz. 95 ListenerM,2kHz (a)(b) (c) Figure2.17SameasFig.2.11butforlistenerMat2kHz. 96 ListenerM,3kHz (a)(b) (c) Figure2.18SameasFig.2.11butforlistenerMat3kHz. 97 ListenerM,4kHz (a)(b) (c) Figure2.19SameasFig.2.11butforlistenerMat4kHz. 98 ListenerL,2kHz (a)(b) (c) Figure2.20SameasFig.2.11butforlistenerLat2kHz. 99 ListenerL,3kHz (a)(b) (c) Figure2.21SameasFig.2.11butforlistenerLat3kHz. 100 ListenerL,4kHz (a)(b) (c) Figure2.22SameasFig.2.11butforlistenerLat4kHz. 101 ListenerV,2kHz (a)(b) (c) Figure2.23SameasFig.2.11butforlistenerVat2kHz. 102 ListenerV,3kHz (a)(b) (c) Figure2.24SameasFig.2.11butforlistenerVat3kHz. 103 ListenerV,4kHz (a)(b) (c) Figure2.25SameasFig.2.11butforlistenerVat4kHz. 104 Resultsofastatisticalanalysisofthechangeinresponsesassociatedwiththepresence ofAMisshowninFigs.2.26{2.28.Thep-valuesarecalculatedfromatwo-tailedt-testwith thenullhypothesisthatresponsestoSAMarethesameasresponsestosinetones.Thep- valuesthattherearemanycaseswheretheAMtrialsproducedtlyt responses( p< 0 : 05).ThisisnotnecessarilyduetotheEITDcue,butmayalsobe byachangeinILD.TheILDisexpectedtochangelittleduetothetheadditionofAM itself.Mostofthechangeisprobablyrandomandcausedbythelistener'sposition.Withthe exceptionofwhenp-valuesverycloseto1.0,caseswithlargerp-valuesdonotnecessarily indicatethattheresponseswerestatisticallyindistinguishable.Thiscouldsimplybethe resultofaninconsistentlistenerresponse.Thep-valuesshouldonlybeexpectedtobesmall insituationswherelistenerresponsesareconsistentandeithertheILDhastly changedand/ortheEITDcueistlywiththeILD.Becausealarge numberoft-testswereperformed,itislikelythatthereareinstancesinwhichthenull hypothesiswasincorrectlyrejected.Therefore,thesep-valuesshouldbeconsideredasa whole,ratherthanplacingemphasisonanyoneinparticular. Thestatisticalceofthechangesinresponsevarygreatlyonaloudspeaker-by- loudspeakerbasis.Fisher'smethod[12]isameta-analysisofthep-valuesbasedonaone- sided ˜ 2 testandshowstheofthechangesinresponseacrossallloudspeakers. Thenull-hypothesisisthatthemeanresponsesforallloudspeakersarethesameinthesine andAMruns.Thealternativehypothesisisthatthemeanresponseforatleastoneofthe loudspeakersistforthesineandAMruns.Inotherwords,thealternativehypothesis isthatlistenersgavetresponseswithAMthanwithsinetones.Atasigni levelof =0 : 05,alloftheFisher'smethodp-valuesrejectthenullhypothesisexceptfor listenerBat4kHz.Again,althoughitisclearthatlistenersgavetresponseinthe 105 p-valuesforchangesinresponseat2kHz Figure2.26p-valuevs.speakernumberforchangesinresponseat2kHz.Eachlistener isplottedaccordingtothesymbolsinthelegend.Thep-valuesarebasedonatwo-tailed t-testandrepresenttheprobabilitythattheresponsesinthesinetoneandAMtrialsare statisticallysimilar.Areferencelineisshownfor =0 : 05. 106 p-valuesforchangesinresponseat3kHz Figure2.27SameasFig.2.26butfor3kHz. 107 p-valuesforchangesinresponseat4kHz Figure2.28SameasFig.2.26butfor4kHz. 108 Fisher'smethodp-values Listener 2kHz 3kHz 4kHz B 4 10 12 4 10 14 0 : 1084 C 8 10 11 8 10 12 5 10 9 M 3 10 7 2 10 13 4 10 6 L 2 10 6 7 10 4 0 : 0151 V 7 10 11 6 10 16 1 10 4 Alllisteners 5 10 39 1 10 49 1 10 15 Alllisteners/frequencies 1 10 97 Table2.3Fisher'smethodp-values.Thismeta-analysisusesaone-sided ˜ 2 testwith2 n degreesoffreedomonthesummationoverthelogarithmsofthet-testp-valuesforthe loudspeakers,where n isthenumberoft-testp-valuesinthesummation.Resultsareshow foreachfrequencyandlisteneroverthe13loudspeakers,foreachfrequencyacrossalllisteners andloudspeakers,andacrossallfrequencies,listeners,andloudspeakers. AMruns,thiscannotnecessarilybefullyattributedtothepresenceoftheEITD,astheILD isknowntohavechangedaswell,evenifonlyslightly. Ameasureoftheaccuracyofthelisteners'responsesisshowninFig.2.29.Thisplotshows thecorrelationsbetweenlistenerresponsemeansandtheactualsourceazimuth.Itdoesnot involvetheinterauralcues.Thecorrelationsrepresenthowwellthelistenerslocalizedthe source. Ineverycase,exceptforlistenerVat4kHz,theAMcorrelationsarehigherthanthesine correlations.Althoughsomeoftheimprovementsaremodest,theintroductionofamplitude modulationvirtuallyneverresultedinpoorerlistenerperformanceatanyofthesefrequencies. ForboththesinetonesandSAMtones,performancetendstoincreaseasfrequencyincreases. ThiscanbeattributedtothefactthethepeakoftheILDcurveadvancestolargerazimuths asthefrequencyincreases,withtheexceptionoflistenersCandVat4kHz.Therefore, fewerloudspeakersinthearrayproducemisleadingILDs.Thisalsoresultsinthecorrelation betweenILDandsourceazimuthincreasingwithfrequency,exceptforlistenersCandLat 109 4kHz(seeFig.2.43).Inotherwords,theILDcuetendstobemorereliableasthefrequency increases.LookingbackatFig.2.29,theAMcorrelationstendtonotincreaseasmuchasthe sinecorrelations.ThismaybeexplainedbytheEITDsolvingtheproblemofthemisleading non-monotonicILDcue.Itiseatallofthesefrequencies.Thiscanalsoexplainwhy theimprovementinperformanceincreasesasthefrequencydecreases(Fig.2.29). Figure2.30showscorrelationsbetweenlistenerresponsesandtheILDforsineandAM runs.Mostcorrelationsarequitehigh.Thisisexpectedtobethecasewiththesinetones. WiththeexceptionoflistenerC,theAMcorrelationsarealsolarge.Theonlycaseswherethe AMcorrelationislargerthanthesinearelistenerBandMat4kHz.TheAMcorrelations areexpectedtobelowersinceresponsesarealsobytheEITD.However,itis strikinghowsmallthecorrelationare.Therearetwopossiblereasonsforthis. TheisthattheEITDmaynothavehadastrongtonlistenerresponses.The secondisthattheEITDandtheAMILDthemselveswerehighlycorrelated.Inthiscase, thecueswouldberedundantandcorrelationswithresponsewouldbehighforeachofthem. TheredbarsinFig.2.31showthiscorrelation.OnecanseethatcaseswheretheAMILD andEITDcorrelationissmallercorrespondstosituationsinFig.2.30withlargerdisparities betweenthetwocorrelations.Forexample,inFig.2.30listenerChasfairlylowcorrelations betweenresponseandAMILD(redbars),whilelistenerLhasveryhighcorrelations.Figure 2.31showsthatlistenerChaslowcorrelationsbetweenthetwocues,whereasforlistener Ltheyarehigh.ThisshowsthateventhoughthecorrelationsbetweenresponseandAM ILDishigh,thattheILDisnotnecessarilysolelycausingtheresponses.TheEITDmust relevanttoo. ThebluebarsinFig.2.31showthesimilaritybetweentheILDinthesineandAMruns. Asstatedpreviouslythesecuesareindeedverysimilar. 110 Response&SourceAzimuth Figure2.29Pearsonproduct-moment(PPM)correlationcotsforperfectresponses (sourcenumber)andactualresponsesareshownforAMinredandsineinblue.Foreach listener,thethreefrequenciesandthemeanandstandarddeviationacrossfrequenciesare shown.Inthelowerrightisthemeanandstandarddeviationacrossalllistenersforeach frequencyandthemeanandstandarddeviationacrossalllistenersandfrequencies.The errorbarsaretwostandarddeviationsinoveralllength. 111 Response&ILD Figure2.30Pearsonproduct-moment(PPM)correlationcotsforAMresponsesand AMILDareinred.CorrelationsforsineresponsesandsineILDareinblue.Foreach listener,thethreefrequenciesandthemeanandstandarddeviationacrossfrequenciesare shown.Inthelowerrightisthemeanandstandarddeviationacrossalllistenersforeach frequencyandthemeanandstandarddeviationacrossalllistenersandfrequencies.The errorbarsaretwostandarddeviationsinoveralllength. 112 Correlationsbetweencues Figure2.31Pearsonproduct-moment(PPM)correlationcotsforAMILDandEITD areinred.CorrelationsforAMILDandsineILDareinblue.Foreachlistener,thethree frequenciesandthemeanandstandarddeviationacrossfrequenciesareshown.Inthelower rightisthemeanandstandarddeviationacrossalllistenersforeachfrequencyandthemean andstandarddeviationacrossalllistenersandfrequencies.Theerrorbarsaretwostandard deviationsinoveralllength. 113 TodeterminewhichcuemaybeplayingalargerroleinSAMlocalizationonecancompare thecorrelationsbetweenresponseandthetwocues,asinFig.2.32.Theseplotsarea summaryofthecorrelationsshowninFigs.2.11{2.25.Althoughnottrueineverycase,the trendisfortheILDcuetohaveahighercorrelationwithresponsesthantheEITD. 2.3.2.2SAMandsine EventhoughtheILDseemstohavemoreweightthantheEITDindetermininglistener responses,onewouldexpectthattheEITDisthecueresponsibleforchanginglistener responseswiththeintroductionofamplitudemodulation.ThisisbecausetheILDhas changedverylittle,whereastheEITDisanew,systematiccue.Totestthis,thecorrelations betweenthechangeinlistenerresponse(AM sine)andthechangeincuewereinvestigated. TheseresultsareshowninFig.2.33. WhilelistenerBandCappeartobestronglybytheintroductionoftheEITD cue,duetothethelargecorrelationsbetweenchangeinresponseandEITD,theotherthree listenersdonot.Becauseofthis,andsincethechangeinILDisessentiallyrandom,one mightexpectthatthatlistenersBandChaveimprovedthemostwiththeintroductionof AM.Figure2.29supportsthishypothesis. Figure2.34showstheofnegativeEITDsonlistenerresponses.TheAMresponse forindividualtrialsminusthemeansine-toneresponseforthesameloudspeakerisplotted vs.EITDforalllistenersandfrequenciesininstanceswheretheEITDisnegative.Aline ofbestisshownforthesedata( r =0 : 23066, r 2 =0 : 053204).Mostofthechanges inresponsearenegative,indicatingthattheEITDcueismostoftencausing listenerstolocalizesoundsourcesclosertothemidline.ThechangeinILD(AM sine)is indicatedbythecolorofthedatapoints.ThenegativechangesinILDappeartocorrespond 114 AMResponse&ILD,EITD Figure2.32Pearsonproduct-moment(PPM)correlationcotsforAMresponsesand AMILDareinred.CorrelationsforAMresponsesandEITDareinblue.Foreachlistener, thethreefrequenciesandthemeanandstandarddeviationacrossfrequenciesareshown.In thelowerrightisthemeanandstandarddeviationacrossalllistenersforeachfrequencyand themeanandstandarddeviationacrossalllistenersandfrequencies.Theerrorbarsaretwo standarddeviationsinoveralllength. 115 Cue&Response(AM sine) Figure2.33Pearsonproduct-moment(PPM)correlationcotsforthechangeinre- sponseandthechangeinILDareshowninred.Correlationcotsforthechangein responsesandtheEITDareshowninblue.Foreachlistener,thethreefrequenciesandthe meanandstandarddeviationacrossfrequenciesareshown.Inthelowerrightisthemean andstandarddeviationacrossalllistenersforeachfrequencyandthemeanandstandard deviationacrossalllistenersandfrequencies.Theerrorbarsaretwostandarddeviationsin overalllength. 116 tosmallerchangesinresponse,andpositivechangesinILDappeartocorrespondtolarger changesinresponse.Therefore,thechangeinILDmayaccountforthecaseswherethe changeinresponsewaspositive,eventhoughtheEITDwasnegative.Figure2.35examines theresiduals,ortheverticaldeviationfromthelineofbestinFig.2.34,vs.theILD. Althoughthereissomescatter,presumablyduetotheinconsistencyoflistenerresponses ( r =0 : 31731, r 2 =0 : 10069)thelineofbestshowsthatthereisatrendthatastheILD increases,sodotheresidualsfromFig.2.34. 117 NegativeEITDonresponse r =0 : 23066 slope=0 : 0021675loudspeakers/ s y-intercept= : 55731loudspeakers Figure2.34Changeinresponse(AM sine(mean))vs.EITDforallnegativeEITD.All listenersandfrequenciesatcombined.TheverticalaxssindicatesAMresponsesforindi- vidualtrialsminusthemeansineresponseforthesameloudspeaker.Thehorizontalaxis indicatestheEITDmeasuredintheindividualAMtrials.Thecorrelationcot| r |, slope,andy-interceptforthebestlineareabovetheplots.Thecolorscaleindicatesthe valueofthechangeinILDbetweentheindividualAMandaveragesineruns.Afewofthe ILDvaluesclipthetopofthecolorscaleat+5dB.Areferencelineforzeroresponse isshown. 118 r =0 : 31731 slope=0 : 17674loudspeakers/dB y-intercept= 0 : 68189loudspeakers Figure2.35ResidualsfromthebestlineinFig.2.34vs.ILD.Alineofbestis shownaswellasreferencelinesthroughtheorigin. 119 2.3.2.3CompressiveCues Figure2.36showsaplotfromanIPDlateralizationexperimentbyYost(1981)[50].For interauralcuesthataresmallinmagnitude,responsescanbewell-approximatedtobelinear. However,inthisplot,onecanseethatafterabout 90 IPD,thelateralizationresponse judgementsbegintosaturate.ThisisalsotrueofILDs[50]andEITDs[3].Saturationshould alsooccurinlocalizationtasks.Perceptualresponsestointerauralcuesarecompressive functions,ofsomekindoranother. Ananalysisthattakesthiscompressivenatureintoaccountmaybetterillustratethe oftheEITDvs.theonrencesinlisteners'changesinresponses.Theprevious analysisofthecorrelationbetweenthechangeinresponseandthechangeininterauralcue wasrepeatedusingcompressedcues.Theamountofcompressionwasvariablebetweenno compressionandextremecompression.Foreachlistener,frequency,andinterauralcue,a particularamountofcompressionwaschosensuchthatthecorrelationbetweenthechange inresponseandcuewasmaximized.Althoughneitherofthecompressionextremesrepresent whatmaybeconsideredtobecorrect,thisanalysiswillshedlightonhowmeaningfuleach ofthecueswereinchanginglistenerresponses.Thecuesaregiventhebestpossiblechance tocorrelatewithresponses.Also,thisservedasachecktodetermineifthistechniqueyields reasonableamountsofcompression. TheEITDwastransformedas EITD ! sgn(EITD) j EITD j x 1 (2.36) wheretheexponent, x 1 ,wasselectedbetween0.01{1.00in0.01incrementsinordertomax- 120 IPDlateralizationdatafromYost,W.A.(1981) Figure2.36FIG.3fromYost(1981).Inthisheadphoneexperiment,listenerslateralized on-goingIPDcuesat500Hz.Thetsymbolsindicatetlisteners,andthe verticallinesindicaterangesofresponsesforIPDsof0 , 90 ,and180 . 121 imizethecorrelation.Similarly,thechangeinILDwastransformedas ! sgn(ILD AM ) j ILD AM j x 2 sgn(ILD sine ) j ILD sine j x 2 (2.37) whereanotherexponent, x 2 ,wasselectedbetween0.01{1.00in0.01incrementsinorderto maximizethecorrelation.Becausethecorrelationsshouldnotbenegative,themaximization selectedthemostpositivecorrelation,ratherthanthecorrelationwiththelargestmagnitude. Thecorrelationsandtheexponents, x 1 and x 2 ,werethenreportedforeachcue. Figures2.37{2.41showthechangesinresponseandcompressedcues.ForlistenerB, showninFig.2.37,theEITDcorrelations(labeledPPM)exceedtheILDcorrelationsat eachfrequency.Onecanseewhy,speakerbyspeaker.Inpanel(a),thetheexponentis0.01, yieldingacorrelationofonly0.5757.Thisamountofcompressionisunreasonable.However, ifatotallychangeinILDyieldedthebestcorrelationwiththechangeinresponse,this changeinILDwasapparentlynotthechangeinlistenerB'sresponseat2kHz. ListenerB'schangesinresponsearetoalargedegreeonlybytheEITDcue. ForlistenerC,showninFig.2.38,theresultsaresimilartothoseoflistenerB.The EITDcorrelationsarefairlylargeandgreaterthanthecorrespondingcorrelationsforthe changeincompressedILD.Onaspeaker-by-speakerbasis,itisconvincingthatthislistener's changesinresponsewherepredominatelybytheintroductionoftheEITDcue.The exponentsfortheEITDcuearereasonable.FortheILD,theexponentsaremuchsmaller, especiallyat3and4kHzwhereitis0.01. ForlistenerM,showninFig.2.39,theoftheEITDislessconvincingthanfor listenersBandC.At2kHz,theEITDcorrelationbeatstheILDcorrelation.ButtheEITD correlationisnotconvincingbyitself.Onlyforspeakers0{4duethechangesinresponse 122 appeartocorrelatewellwiththeEITD.HowevertheILDcorrelationisnegative,whichis totallyunreasonable.Noticethoughthattheresponsesdidnotchangedrastically.Because ofthis,alargecorrelationisunlikelytoappearanywhere.Again,at3kHz,theEITD correlationbeatstheILDcorrelation,butisnotconvincingbyitself.Butatboth2and 3kHz,theEITDsappeartobelargeenoughthatonewouldexpectalargerchangein response.ThislistenerseemstobesomewhatinsensitivetotheintroductionofEITDand islargelybasinghisresponseintheexistingILD(whichhasnotchangeddrastically).At 4kHz,theILDcorrelationbeattheEITDcorrelation,butneitherofthemareconvincingly large. ForlistenerL,showninFig.2.40,theEITDdoesnothaveaconvincingonthe changesinresponse.At2kHz,thechangeinILDcorrelatesfairlywellwiththechangein response.TheEITDcorrelationisquitealotsmaller,andismaximizedwhentheEITD istotallycompressed.At3kHz,neitherthechangeinILDnortheEITDarestrongly correlatedwiththechangeinresponse.Infacttheresponsesdidnotchangemuchatallat 3kHz.Thismakesitunlikelythattherewouldbeahighcorrelationforeithercue.Thesame istrueat4kHz.Atboth3and4kHztheEITDcueappearsasifitshouldbeextremely helpful,asittendstoriseroughlylinearly.ButtheresponsechangeifmostlyThis maybebecauselistenerLisinsensitivetotheEITD,andpreferstheILDcue.Orpossibly, thislistener'ssine-toneresponseswerealreadyquitegoodandcouldnotbeimprovedupon. Lookingatthemeanresponsesfor3and4kHzinFigs.2.21and2.22,respectively,shows thatthereisroomforthislistenertoimprove.At3kHz,themeanresponsedidnotexceed speaker9andat4kHzthemeanresponseslightlyexceededspeaker10.Orperhapsthis listener'sinexperienceinpsychoacousticalexperimentscontributedtoanapparentlackof EITDsensitivity. 123 ForlistenerV,showninFig.2.41,thechangesinresponsearenotverydrastic.At2kHz, neithercorrelationsareverystrong.At3and4kHz,theEITDcorrelationsarenegative, whichessentiallyindicatesthatthelistener'sresponsewasnotectedbytheEITD.Like listenerL,listenerV'sresponseshadroomforimprovement,asseeninFigs.2.23{2.25. Figure2.42isasummaryofthecorrelationsfromFigs.2.11{2.25.ItissimilartoFig. 2.33,exceptthatthecorrelationshavebeenmaximizedbythecompressed-cuesanalysis. Unfortunately,thecompressed-cuesanalysisdidnotresultindrasticallyrentcorrelations thantheuncompressed-cuesanalysis.Naturally,thelargestdiscrepanciesappearwhenthere isalotofcompression.Thecompressionexponentstendedtobeverynearoneorzero. Althoughtheexponentsnearzeroyieldedthelargestcorrelations,theyaretotallyunrealistic, andindicatethatchangeinresponseandcuehaveweakcorrelation.Noticethatthecases withverysmallcompressionexponentsoccurfortheforlistenersBandCandfor theEITDforlistenersLandV.Thiswouldindicatethatintheseinstances,thelisteners' changesinresponsearebeingdictatedalmostentirelybytheoppositecues.Thisstatement agreeswiththegeneraltrendofthecorrelationsfortheselisteners. Therencesbetweenindividualsherehasacoupleofpossibleexplanations.The possibilityisthatthereareindividualesintherelativesensitivitytotheILDand EITDcues.ListenersBandCappeartobestronglybytheEITD.Butlisteners M,L,andVdonot,tovaryingdegrees.tlistenersmayhavetauditory weightingfortcues.However,thereisanotherpossibility.Considerthefactthat theseelistenersdidnotreceivethesamesignalsintheirearcanals.Thiswouldbedue mostlytointheexternalanatomyofthetlisteners.Itcouldbethat listenersM,L,andVdidnotgetanyinformationfromtheEITDthatcouldhavebeenof anyuse.IftheILDwasalreadyquitelargeandproducedaresponsewithalargeazimuth, 124 thenthereisnotmuchroomforimprovementbyanadditionallargeEITDcue.TheEITD cuesthatlistenersM,L,andVreceivedmaynothaveverymuchwiththeILD thatwasalreadypresent. 2.3.2.4SAMinterauralcuereliability AwaytounderstandtheinperformancechangeshowninFig.2.29istolookat thereliabilityofthecues.Figure2.29showsthegreatestimprovementsforlistenersBand C.DidtheselistenershavesubstantiallymorereliableEITDcuesthanILDcues?Figure 2.43showsthecorrelationbetweenthetwoAMcuesandtheazimuth.Inprinciple,the largerthesecorrelationsare,themorereliabletheyareforlocalization. IfcuereliabilityisthecauseofindividualthenlistenersBandCshould showntcuereliabilitythantheotherlisteners|especiallylistenersLandV.However, listenersBandCdon'tshowmorereliableEITDcuesthantheotherlisteners. AnargumentcouldbemadethatlistenersBandChavelessreliableILDcuesthanthe otherlisteners.Thiscouldsupportthehypothesisthattheselistenershadmoretogainby listeningtotheEITDcuethantheotherlisteners.Thismaythenexplainwhylisteners BandCshowedmorecorrelationwiththeEITDandtheirchangeinresponsesinFigs. 2.33and2.42.Furthermore,Fig.2.31indicatesthatlistenersBandChadmorecasesof dissimilarcues,thanforlistenersL. 2.3.3Discussion Itremainssomewhatunclearsofarastowhythereareindividualdiinthethe correlationsbetweenlisteners'changesinresponseandthecuesshowninFigs.2.33and2.42. Butthereisnotoverwhelmingevidencetosuggestthatthecuesthemselvesareresponsible 125 CompressedCues,ListenerB (a)(b) (c)(d) (e)(f) Figure2.37ChangesinresponseandcompressedinterauralcuesforlistenerB.The rowis2kHz,thesecondrowis3kHz,andthethirdrowis4kHz.Foreachplot,redcircles indicatechangesinmeanresponse(AM sine)anderrorbarsaretwostandarddeviations inoveralllength.Forplots(a),(c),and(e),bluetrianglesindicatethechangeincom- pressedILD(AM sine).Forplots(b),(d),and(f),bluediamondsindicatethecompressed EITD.Foreachplot,themaximizedPearsonproduct-moment(PPM)correlationcoet betweenthechangesinresponseandcompressedinterauralcueisshown,aswellasthe compressionexponent. 126 CompressedCues,ListenerC (a)(b) (c)(d) (e)(f) Figure2.38SameasFig.2.37butforlistenerC. 127 CompressedCues,ListenerM (a)(b) (c)(d) (e)(f) Figure2.39SameasFig.2.37butforlistenerM. 128 CompressedCues,ListenerL (a)(b) (c)(d) (e)(f) Figure2.40SameasFig.2.37butforlistenerL. 129 CompressedCues,ListenerV (a)(b) (c)(d) (e)(f) Figure2.41SameasFig.2.37butforlistenerV. 130 CompressedCue&Response(AM sine) Figure2.42MaximizedPearsonproduct-moment(PPM)correlationcotsforthe changeinAMandsineresponsesandthechangeincompressedILDareshowninred. CorrelationcotsforthechangeinAMandsineresponsesandthecompressedEITD areshowninblue.Thegreencirclesindicatethecompressionexponent.Foreachlistener, thethreefrequenciesandthemeanandstandarddeviationacrossfrequenciesareshown.In thelowerrightisthemeanandstandarddeviationacrossalllistenersforeachfrequencyand themeanandstandarddeviationacrossalllistenersandfrequencies.Theerrorbarsaretwo standarddeviationsinoveralllength. 131 AMILD,EITD&SourceAzimuth Figure2.43Pearsonproduct-moment(PPM)correlationcotsforAMtones.Corre- lationsbetweenILDandsourceazimuthareinred.CorrelationsbetweenEITDandsource azimuthareinblue.Foreachlistener,thethreefrequenciesandthemeanandstandard deviationacrossfrequenciesareshown.Inthelowerrightisthemeanandstandarddevia- tionacrossalllistenersforeachfrequencyandthemeanandstandarddeviationacrossall listenersandfrequencies.Theerrorbarsaretwostandarddeviationsinoveralllength. 132 fortheindividualItisentirelypossiblethatlistenersareprocessingILDand EITDtly. WhatisshownbyFig.2.29isthataddingamplitudemodulationnearlyneverhurts listeners,andissometimesbEventhoughthereliabilityoftheEITDcueitselfis oftenwed,becausetheILDcuefromnon-monotonicity,andbecausetheEITDcue tendstobestrongestafterthepeakoftheILDcurve,itiseasilyunderstandablehowAMis almostalwaysb 2.3.4AMQuality(Problem1) ThekeytounderstandingindividualmaylieinanalyzinghowAMquality(prob- lem1)listenerresponses[1].PerhapslistenersBandChadhigherAMquality,and werethereforebetterabletoutilizetheEITDcuethantheotherlisteners.AMqualitywas analyzedinseveralways:First,theof m andenvelopecoherenceonthechangein responseforloudspeakers8{12.Thenthe m andenvelopecoherenceonthestrength ofEITD.Finally,theof m andenvelopecoherenceontheresponseerrorratio. 2.3.4.1Changeinresponsevs.AMquality BecausetheEITDshouldbemostusefulatlargeazimuths(pastthepeakoftheILD),this analysiswillberestrictedtothatregion.Theseanglescorrespondtoabout60 to90 ,or speakers8{12.HeretheEITDislarge,soifitsqualityisreduced,asmallpercentagechange inlistenerlocalizationresponseshouldbeeasiertomeasurethanatsmallerangleswhere therearefewerspeakersnearthemidlineforalistenertorespondto.Alsoattheselarge azimuths,theEITDandILDbegintocontradicteachother,sotheroleoftheEITDon responseshouldbemorepronounced. 133 Figures2.44and2.45showhowresponsechanged(AM sine)fortheleft(far)earand right(near)ear,respectively.Theleftearhasamuchlargerspreadin m thantherightear, sotheAMquality'sonitshouldstronger(ifatall).Howeverthereseemstobevery little,ifany,systematictrendinthesedataforeitherearasthecorrelationsfortheleftand rightearsare r = 0 : 1031and r = 0 : 0973,respectively.Thismightbebecauseallofthe m valuesherearetoohightohindertheabilitytouseEITD.Thelowestvaluefortheleft earisabout0 : 75,andfortherightearisnotmuchlessthan1.Althoughover-modulationis saidtobeAMoflowerqualitythantheSAMsignal,itispossiblethatinsomecircumstances itisatleastasusefultolistenersas100%AM.However,thisexperimentwasnotdesigned torigorouslycoveraparticularrangeof m values.Ratheritisarepresentationofwhat alistenerinfreewouldencounterwhenlisteningtoSAMtonesatvariousazimuths. Although m iscertainlyinfreeforbothears,itmaybethatnotrendcanbe seenbecause m simplyisalwaystoohigh.Infact,NuetzelandHafter(1981)[40]showed thatfor4-kHzSAMtonesmodulatedat150and300Hz,the(JND) intheEITDdoesnotvarygreatlywith m when m isgreaterthanabout0.5.TheJND increasessuddenlyas m decreasesbelowabout0.4.Thevalueof m isgreaterthan0.5for speakers8{12,asseeninFigs.2.44and2.45,andthisremainstrueacrossallloudspeakers asseeninthehistogramfromFig.2.8. Anotherhypothesisisthattheinterauralenvelopecoherence|thepeakofequation 2.35|shouldlistenerresponses.Sinceenvelopesintheleftandrightearmustbe comparedduringtheprocessingofEITD,onemightexpectthatiftheenvelopesaretoo dissimilar,theEITDwouldnotbeabletobedetectedaswell.Figure2.46showshowre- sponsechanged(AM sine)vs.theenvelopecoherenceforspeakers8{12.Surprisingly,the envelopecoherenceisalwaysprettyhigh|usuallyhigherthan0 : 99.Thelowestenvelope 134 Changeinresponsevs. m -leftforspeakers8{12 Figure2.44Changeinresponsevs. m -leftear.Allvelistenersareshownhere.Eachpoint representsthemeanchangeinresponseacross10trialsvs.themean m across20intervals forthelistener'sleft(far)ear.Thecorrelationis r = 0 : 1031and r 2 =0 : 01063. 135 Changeinresponsevs. m -rightforspeakers8{12 Figure2.45Changeinresponsevs. m -rightear.Allelistenersareshownhere.Each pointrepresentsthemeanchangeinresponseacross10trialsvs.themean m across20 intervalsforthelistener'sright(near)ear.Thecorrelationis r = 0 : 0973and r 2 =0 : 00947. 136 Changeinresponsevs.interauralenvelopecoherenceforspeakers8{12 Figure2.46Changeinresponsevs.interauralenvelopecoherence.Allelistenersare shownhere.Eachpointrepresentsthemeanchangeinresponseacross10trialsvs.themean m across20intervals.Thecorrelationis r = 0 : 0633and r 2 =0 : 00440. coherenceisnolessthan0 : 95.Thisisdespitethepresenceofunusualpairsofwaveforms suchasthoseinFigs.2.3and2.4.Thecorrelationis r = 0 : 0633.Unfortunately,justas wasthecasewith m ,itmaybethatthisparticularexperimentdidnotresultinanenvelope coherencewhichwaslowenoughtoshowasystematiconthechangeinresponse. Inantosomeseveralvariationsonthisanalysiswereconducted.Table 2.4showsasummaryofthesecorrelations.Anotherhypothesisisthat m =1istheideal 137 signal,andthatanydeviationfromthiswillcauseproblemsforalistener.So,all m< 1 werefoldedover1suchthat m folded =1+(1 m ).Thiswasdoneforbothears.Inboth casesthecorrelationsbecameevenweaker,andnopatternwasnoticeableonthescatter plots.Thenexthypothesiswasthatthesmallest m (leftorright)iswhatmatters.This producedthestrongestcorrelationofall, r = 0 : 2435.However,theexpectationisthatas thesmallest m increases,thattheEITDcueshouldbemoreeatincreasinglistener responsesforspeakers8{12.Thenegativecorrelationhereopposesthishypothesis.Yet sincethiscorrelationisstillquitelow,thishypothesisshouldnotbediscounted.Finally, the m (leftorright)whichwasfurthestfrom1wasanalyzed.This,too,resultedinaweak correlation. r forspeakers8{12 independentvariable r m -leftear -0.1031 m -rightear -0.0973 envelopecoherence -0.0633 m -left,foldedover1 0.0043 m -right,foldedover1 -0.0387 smallest m (leftorright) -0.2435 m ,furthestfrom1(leftorright) -0.0917 Table2.4Thecorrelationcots, r ,forFigs.2.44{2.46andfourothertests.Thedata includealllistenersandallfrequencies.Foreachloudspeaker,thedataarethechangein meanresponses(AM sine)vs.themeanindependentvariable.For m -leftand m -rightfolded over1,valuesof m< 1aretransformedtovaluesof m> 1whilekeeping j m 1 j thesame. 138 2.3.4.2TheofAMqualityoncuestrength TheprevioussectiondidnottakeintoaccountthevalueoftheEITD.Italsoonlylooked atthemeanforeachloudspeaker,andgroupedalllistenersandcarrierfrequenciestogether. Itmaybepossibletocaseswhere m ortheenvelopecoherencethestrength oftheEITDonamoreindividualizedbasis.InFigs.2.47{2.58thechangeinresponse (individualtrialminusmeansineresponse)isplottedvs.theEITD.Thecolorofeachdata pointrepresentseitherthe m fortheleftorrightears,ortheenvelopecoherence.Thegoal istoobservesituationswheredatapointsdeviatefromthebestlines,andtolookfor systematicMostoftheshouldcoincidewiththelargestvaluesofEITD.Let thecue-weighthypothesisbethattrialswithlowerqualityAMwilltendtofallbelowthe lineofbestandtrialswithhigherqualityAMwilltendtofallabovethelineofbest Additionally,thisshouldincreasewithEITD.Forpanels(a)and(b)thecue-weight hypothesisisthatunder-modulatedtoneswilltendtofallbelowthebestline,and100%- modulatedorover-modulatedtoneswillfallonorperhapsabovethelineofbestFor panels(c),thecue-weighthypothesisisthatlowercoherencestimuliwilltendtofallbelow thelineofbestandhighercoherencestimuliwilltendtofallonorabovethelineofbest InFig.2.47thechangeinresponseincreasessuddenlyaround750 s.Theleftear(far ear),though,showsthatthesehappenedtobesomeofthemostunder-modulatedtonesin theseruns.Theresponsesincreaseddramaticallydespitebeingsomewhatunder-modulated inthefarear.Thisdoesnotmeanthattheunder-modulationcausedthis,butitdoesdoes opposethecue-weighthypothesisinthiscase.Therightearshowsamixtureofmodulation valuesathighEITD|somelessthanzeroandsomegreater.Theenvelopecoherenceislarge 139 inmostcasesandnoobvioustrendisvisibleinthisplot. InFig.2.48theslopeofthebestlinehasdecreased,comparedforFig.2.47.This indicatesthattheEITDcuewasnotasimpactfulat3kHzasitwasat2kHz.Thisislikely becausetheILDcueismorereliableat3than2kHz,sothesineresponsesaresomewhat moresaturatedandthereislessroomforimprovement.Fortheleftear(panela)theredoes appeartobesomeconsistencyinhow m isdistributedontheplot.Sincetherearetentrials perloudspeaker,itisnotsurprisingtoseethattherearesimilargroupingsofEITDand m . Buttheredoesnotappeartobeasystematicdistributionofover-modulatedtonesabove thebestlineandunder-modulatedtonesbelowtheline.Therightear(panelb)shows m isusuallyclosertounitythantheleftear.Again,though,thecue-weighthypothesisisnot supportedhere.Theenvelopecoherence(panelc)isalwaysveryhighandnoonthe changeinresponsecanbeseen.ForlistenerC,theslopeinFigs.2.50{2.52alsodecreases withincreasingcarrierfrequency.Aninspectionoftheseplotsshowsnosupportforthe cue-weighthypothesis. InFig.2.49,theslopeofthebestlinecontinuestodecreasecomparedtothethelower carrierfrequenciesinFigs.2.47and2.48.Onceagaintheleftearhasawiderrangeof m thantheleft.Nocleartrendisnoticeablefor m -leftear, m -rightear,ortheenvelope coherence. ForlistenerM,showninFigs.2.53{2.55,theslopesaresmallerthanforlistenersBand C,whichisconsistentwiththefactthatlistenerMwaslessbytheEITD.Even so,itmaybeinterestingtoseeif m ortheenvelopecoherenceplayedarollinthechange inresponse.Therearesometrendsinhow m isdistributedalongthehorizontalaxis|for example,Fig.2.17,panel(a).Butagain,thisrelates m tothevaluesoftheEITDitselfand thereforeroughlytotheazimuth.Whatisnotseenisthatforover-modulatedtones,changes 140 ListenerB,2kHz r =0 : 65247 slope=0 : 0056987loudspeakers s y intercept= 1 : 0589loudspeakers (a)(b) ˙ m left =0 : 64318 ˙ m right =0 : 13722 (c) ˙ envelopecoherence =0 : 018504 Figure2.47Changesinresponsevs.EITDwithAMqualityforlistenerBat2kHz.All plotsdisplaythesamedataandintheAMqualitymetricshownincolor.Thevertical axisindicatesAMresponsesforindividualtrialsminusthemeansineresponseforthesame loudspeaker.ThehorizontalaxisindicatestheEITDmeasuredintheindividualAMtrials. Thecorrelationcoet| r |,slope,andy-interceptforthebestlineareabovethe plots.Thecolorscaleinplot(a)indicatesthevalueof m intheleftearintheindividualAM trials.Thecolorscaleinplot(b)indicatesthevalueof m intherightearintheindividual AMtrials.Thecolorscaleinplot(c)indicatesthevalueoftheenvelopecoherenceinthein individualAMtrials.Thestandarddeviations, ˙ ,oftheAMqualityareshownabovethe plots. 141 ListenerB,3kHz r =0 : 73616 slope=0 : 0031456loudspeakers s y intercept= 0 : 21311loudspeakers (a)(b) ˙ m left =0 : 15447 ˙ m right =0 : 051902 (c) ˙ envelopecoherence =0 : 0036517 Figure2.48SameasFig.2.47butforlistenerBat3kHz. 142 ListenerB,4kHz r =0 : 16532 slope=0 : 0005914loudspeakers s y intercept=0 : 27188loudspeakers (a)(b) ˙ m left =0 : 28026 ˙ m right =0 : 065509 (c) ˙ envelopecoherence =0 : 0063304 Figure2.49SameasFig.2.47butforlistenerBat4kHz. 143 ListenerC,2kHz r =0 : 71628 slope=0 : 0046531loudspeakers s y intercept= 0 : 66024loudspeakers (a)(b) ˙ m left =0 : 16724 ˙ m right =0 : 070361 (c) ˙ envelopecoherence =0 : 0036442 Figure2.50SameasFig.2.47butforlistenerCat2kHz. 144 ListenerC,3kHz r =0 : 64037 slope=0 : 0035973loudspeakers s y intercept= 1 : 3085loudspeakers (a)(b) ˙ m left =0 : 12719 ˙ m right =0 : 079148 (c) ˙ envelopecoherence =0 : 0031154 Figure2.51SameasFig.2.47butforlistenerCat3kHz. 145 ListenerC,4kHz r =0 : 74707 slope=0 : 0032856loudspeakers s y intercept= 1 : 9217loudspeakers (a)(b) ˙ m left =0 : 075608 ˙ m right =0 : 05195 (c) ˙ envelopecoherence =0 : 0017147 Figure2.52SameasFig.2.47butforlistenerCat4kHz. 146 inresponsegrowabovethelineofbestwithincreasingEITD,andforunder-modulated tones,changeinresponsegrowbelowthelineofbestwithincreasingEITD.Similarly, thereisnotrendseenwiththeenvelopecoherence.ForlistenersL,showninFigs2.56{2.58, theslopesaresmalllikeforlistenerM.Unfortunately,thecue-weight-hypothesiscannotbe byinspectingtheseplots. ForlistenerV,showninFigs.2.59{2.61,thescatterplotsarethemostunusual.The 2kHzplotsshowninFig.2.59aretheonlyoneswithareasonablylargeslope,meaning thattheEITDischangingthislistener'sresponses.At3kHztheslopeisnegative.Onecan seewhyperhapstheslopewasnegativeinthe3kHzcase.Figure2.24showstheILDfor thesinetonesinpanel(a)andtheAMtonesinpanel(b).Inthisinstance,theAMILDis noticeablysmallerandappearstobedrivingmanyofthechangesinresponsetobenegative. That'snottosaythattheEITDhadbeentotallyignored.ThecorrelationbetweenILDand responsewas0 : 9458forthesinetonesand0 : 8832fortheAMtones,indicatingthatanother variablewasinplay.For4kHz,showninFig.2.61,theslopeisveryThisisnotas easytoexplainbycomparingtheplotsinFig.2.25.However,asisthecasefor3kHz,the correlationbetweenILDandresponseislowerfortheAMtonesinpanel(b)thanforthe sinetonesinpanel(a).SotheEITDissurelynotbeingignoredbylistenerV,butitcannot beshowntobeweightedby m ineitherearnorbytheenvelopecoherence. 147 ListenerM,2kHz r =0 : 3606 slope=0 : 00099731loudspeakers s y intercept= 1 : 7189loudspeakers (a)(b) ˙ m left =0 : 28269 ˙ m right =0 : 094618 (c) ˙ envelopecoherence =0 : 010209 Figure2.53SameasFig.2.47butforlistenerMat2kHz. 148 ListenerM,3kHz r =0 : 28409 slope=0 : 0010331loudspeakers s y intercept= 0 : 16619loudspeakers (a)(b) ˙ m left =0 : 34619 ˙ m right =0 : 09864 (c) ˙ envelopecoherence =0 : 020431 Figure2.54SameasFig.2.47butforlistenerMat3kHz. 149 ListenerM,4kHz r =0 : 1901 slope=0 : 00077117loudspeakers s y intercept= 0 : 98598loudspeakers (a)(b) ˙ m left =0 : 107 ˙ m right =0 : 083703 (c) ˙ envelopecoherence =0 : 0019765 Figure2.55SameasFig.2.47butforlistenerMat4kHz. 150 ListenerL,2kHz r =0 : 10746 slope=0 : 00053605loudspeakers s y intercept= 0 : 44355loudspeakers (a)(b) ˙ m left =0 : 16412 ˙ m right =0 : 11754 (c) ˙ envelopecoherence =0 : 00624 Figure2.56SameasFig.2.47butforlistenerLat2kHz. 151 ListenerL,3kHz r =0 : 1918 slope=0 : 00080706loudspeakers s y intercept=0 : 076306loudspeakers (a)(b) ˙ m left =0 : 13935 ˙ m right =0 : 088844 (c) ˙ envelopecoherence =0 : 001678 Figure2.57SameasFig.2.47butforlistenerLat3kHz. 152 ListenerL,4kHz r =0 : 33551 slope=0 : 0011344loudspeakers s y intercept= 0 : 36495loudspeakers (a)(b) ˙ m left =0 : 12436 ˙ m right =0 : 060235 (c) ˙ envelopecoherence =0 : 0021669 Figure2.58SameasFig.2.47butforlistenerLat4kHz. 153 ListenerV,2kHz r =0 : 5001 slope=0 : 0012707loudspeakers s y intercept= 0 : 62406loudspeakers (a)(b) ˙ m left =0 : 1341 ˙ m right =0 : 1664 (c) ˙ envelopecoherence =0 : 0065084 Figure2.59SameasFig.2.47butforlistenerVat2kHz. 154 ListenerV,3kHz r = 0 : 23917 slope= 0 : 00073934loudspeakers s y intercept= 0 : 91268loudspeakers (a)(b) ˙ m left =0 : 162 ˙ m right =0 : 068044 (c) ˙ envelopecoherence =0 : 0023472 Figure2.60SameasFig.2.47butforlistenerVat3kHz. 155 ListenerV,4kHz r =0 : 049125 slope=0 : 00014484loudspeakers s y intercept= 0 : 37229loudspeakers (a)(b) ˙ m left =0 : 17891 ˙ m right =0 : 0455 (c) ˙ envelopecoherence =0 : 0036016 Figure2.61SameasFig.2.47butforlistenerVat4kHz. 156 Toinvestigatefurtherhow m ineitherearortheenvelopecoherencemaythe strengthoftheEITDcue,ananalysisoftheverticaldeviation(residuals)fromthebest linesinFigs.2.47{2.61wasconducted.AnexampleforlistenerBat2kHzisshownin Fig.2.62.Inthiscase,againthereislittleevidencethatthisrangeof m -valuesis thechangeinresponse,asthecorrelationis r = 0 : 064.Table2.5isasummaryofthe correlationsforeachlistenerandfrequencyfor m -left, m -right,andtheenvelopecoherence.If anything,theanticipationisthatthecorrelationswillbepositive,becausemostdramatically, theEITDcueshouldbepullinglisteners'responsestotherightforazimuthspastthepeak oftheILDcurves.However,thereareaboutasmanynegativecorrelationsasthereare positive,noneofwhichareverylarge.ThelargestcorrelationisforlistenerVat4kHz forthe m intheleft(far)ear.ThiswouldindicatethattheEITDtendedtodecreasethis listener'sresponses,andthatperhapswhenthefarearwasunder-modulated,theEITDhad lessweightthanwhenthemodulationwashigher.Eveninthiscase,however,thestatistics arenotveryconvincing. Apossibleexplanationforthisisthatthevaluesof m wereneverreallythatlow.Itmay bethatfortheseparticularAMsignalsinthat m neverbecomesdegradedtothe extentthattheEITD'sweightbecomestlycompromised.Onemightexpectthat ifthemodulationfrequencywerelarger,andthereforethespectrumwerewider,thatthere wouldbemorevariationinthevaluesof m ,forthefarearinparticular.Similarly,ina room,withmanyctions,onemayexpectthattheamplitudesandphasesasmeasured inalistener'searcanalswouldberandomizedtoagreaterextent,andthatthiswouldresult insmallervaluesof m forbothears[37]. 157 ListenerB 2kHz Figure2.62ResidualsfromthebestlineinFig.2.47vs. m -left.Thecorrelationis r = 0 : 064and r 2 =0 : 0041. 158 r forresponseresidualsvs.AMquality Listenerandfrequency(kHz) m -left m -right envelopecoherence ListenerB 2 0 : 064318 0 : 22606 0 : 062472 3 0 : 040750 0 : 10755 0 : 32536 4 0 : 24596 0 : 13275 0 : 23071 ListenerC 2 0 : 11526 0 : 099286 0 : 18409 3 0 : 027794 0 : 19056 0 : 24153 4 0 : 021158 0 : 45753 0 : 20442 ListenerM 2 0 : 011531 0 : 013131 0 : 041615 3 0 : 028100 0 : 15645 0 : 0064281 4 0 : 095850 0 : 098704 0 : 12751 ListenerL 2 0 : 35295 0 : 15010 0 : 25949 3 0 : 22814 0 : 011681 0 : 17502 4 0 : 27120 0 : 011603 0 : 13470 ListenerV 2 0 : 12742 0 : 054720 0 : 16777 3 0 : 22956 0 : 010630 0 : 023282 4 0 : 41962 0 : 21445 0 : 26906 Table2.5Pearsonproduct-moments, r ,fortheresponseresidualsfromblinesvs. theAMqualitymetricshownbythecolorscaleinFigs.2.47{2.61.Thecorrelationsare calculatedforeachlistener,frequency,andAMqualitymetric. 2.3.4.3Responseerrorratiovs.AMquality DespitethelackofevidencethatthequalityoftheAMwastheweightoftheEITD, itmaybepossibletodetermineifthequalityoftheAMresultedinmoreaccurateresponses. Toexaminethisthermsoftheresponseerrorwasexaminedforthemeanresponses.This wascalculatedforeachlistenerandeachfrequency.Theratioofrmserrorforthesinetones andtheAMtoneswasthenplottedvs. m foreachearandtheenvelopecoherenceaveraged acrossspeakers.Fivelistenersandthreefrequenciesmeansthatthereareonly15datapoints oneachplot. Figure2.63showsthermsratiovs. m -left.Largerrmsratioscorrespondtoimproved 159 listenerperformancewithAM.Thecorrelationispositive,butnotstrong( r =0 : 1160).If thedatapointintheupperleftisignored(listenerCat4kHz),thecorrelationwouldbe stronger( r =0 : 4829),butitishardtojustifydoingso.Figure2.64showsthesamerms errorratiosplottedvs.the m intherightear.Herethecorrelationisnegative,butnot strong( r = 0 : 1459).AlthoughitisknownthatAMdidhelplistenersasseeninFig.2.29, thisanalysisdoesnotshowthatthedepthofmodulationwasimportant,atleastforthese stimuli. Onemightthinkthatthesimilaritybetweentheenvelopesinthetwoearswouldimpact theaccuracyoflistenerresponses.Figure2.65showsthesamermserrorratiosplottedvs.the envelopecoherence.Herethecorrelationisverysmall( r = 0 : 0221).Notrendinresponse accuracycanbeseenaswithincreasingenvelopecoherence.Theenvelopecoherenceshere happentobeprettylarge,andmaybeaboveathresholdnecessarytogivestrongweightto theEITDcue. 2.3.4.4Discussion TheslopesofthebestlinesinFigs.2.47{2.61showthattheEITDappearstobecome lesseasthecarrierfrequencyincreases,butthisverylikelyoccursbecausetheILD becomesmoree.ThequalityoftheAMcannotbeshowntoimpactlistenerresponses inthisexperiment.Howeveritcannotbediscountedeither.Thesamplesizewasnotlarge. Theremayhavebeenasmallhiddeninthenoiseofthedata.Suchistheproblemwith anexperimentinfreeastheexperimenterdoesnotdirectlycontrolthenatureofthe signalsinthelistener'searcanals.TheILD,EITD,andqualityofAMaredictatedbythe interactionofthesoundwaveincidentonthelistener'sanatomy.Acontrolledheadphone experimentwithawiderdistributionofcuesandAMqualitymayyieldtresults. 160 rmserrorratiovs. m -left Figure2.63TheratiooftheerrorinresponsermsforsinetonestoAMtonesvs. m -left averagedacrossallspeakers.Theelistenersandthreefrequenciesareallrepresented.The correlationis, r =0 : 1160and r 2 =0 : 01346. 161 rmserrorratiovs. m -right Figure2.64TheratiooftheerrorinresponsermsforsinetonestoAMtonesvs. m -right averagedacrossallspeakers.Theelistenersandthreefrequenciesareallrepresented.The correlationis, r = 0 : 1459and r 2 =0 : 02129. 162 rmserrorratiovs.envelopecoherence Figure2.65TheratiooftheerrorinresponsermsforsinetonestoAMtonesvs.the envelopecoherenceaveragedacrossallspeakers.Theelistenersandthreefrequenciesare allrepresented.Thecorrelationis, r = 0 : 0221and r 2 =0 : 000488. 163 2.4Conclusion Animportantcueforlocalizingsoundsourcesinthehorizontalplaneistheinterauraltime (ITD).TheITDistheinthetimeofarrivalofasoundwavebetween thetwoears.Whenasoundsourceislocateddirectlyinfrontofalistener(0 azimuth), theearsarelocatedthesamedistancetothesource,sotheITDiszero.Ifasoundsourceis locatedtotherightofalisteneratsomepositiveangle,thenthelistener'srighteariscloser tothesoundsourcethanthelistener'sleftear,sothesoundwavewillarriveattherightear beforetheleftear. Forfrequencieslowerthan1450Hz,humansareabletousetheITDinthestructure ofasoundwave[7].Athigherfrequencies,humansdonothavene-structureITDsensitivity. However,theinterauraltimeintheenvelope(EITD)ofasoundwavecanbeused asalocalizationcue.PsychoacousticalexperimentshavepreviouslystudiedtheEITDcue forlistenerswearingheadphones.Atypicalsignalusedinheadphoneexperimentsthatstudy theEITDisasinusoidallyamplitudemodulated(SAM)sinetone. SAMlocalizationinfreefromlateralizationexperimentswithheadphones. Astheincidentsoundwaveinteractswithalistener'sanatomy,theamplitudeandphase spectrumoftheoriginalSAMsignalisaltered.Mostnotably,thegroupdelayissuchthat theEITDcuebecomessomewhatunreliabletolisteners.Itoftenhastheoppositesignas theazimuth,anddoesnotincreasemonotonically. Therewereinter-listenerinperformanceinlocalizingSAMtones.Apossible reasonforthisisthatthereareindividualencesinthequalityofthecues.Another possibilityisthatthereareindividualintheprocessingoftheinterauralcues. TherewasnotanyevidencefoundlinkingtheAMqualitytolistenerperformance.Therefore, 164 thissuggeststhattheremaybeindividualincueprocessing. Nevertheless,theintroductionofAMalmostalwayshelpedwithlocalizationaccuracy andvirtuallyneverhinderedit.Thisislikelyduetotheinteraurallevel(ILD) cueitselfbeinganunreliablecueduetotheacousticalbrightspot.TheEITDcue,although wed,makesupforthewsintheILDcue,especiallyatlargeazimuths. 165 APPENDIX 166 Appendix SAMtonesinrooms Methods InordertounderstandthethatroomshaveontheinterauralpropertiesonSAM tones,aKEMARmanakinwasusedtomakebinauralrecordingsofSAMtonesinrooms. Tworoomswereused:areverberationroomandaroomcalledthe\laboratory"[18,19,43]. Thereverberationroom(IndustrialAcousticsCompany)was25by21ftwitha12-ftceiling (7.7by6.4by3.6m).Thereverberationroomwasemptyexceptforafewobjectswith minimalsoundabsorption.Thelaboratorywas28by24witha15-ftceiling(8.5by7.3 by4.6m).Theroomhadconcrete-blockwalls,aconcreteceiling,andvinyltile.Todampen theusuallylivelyroom,therewasacanvastarpaulinontheorwithdimensionsof16by 10ft.Additionally,scatteredontheorwereabouttwodozensoundabsorbingacoustical cones. Inthereverberationroom,theKEMARwaspositionedinthreeseparatelocations.At eachlocationtheKEMARwasrotatedtofourtangles:0,30,60,and90degrees. AtpositiveanglestheKEMARwasrotatedsuchthattheloudspeakerwasontherightside. Ateachangletheloudspeakerwasplacedat3distances, d :50,100,and200cm.Location1 was9ft2inintotheroomalongaline10ft9ininfromtheright-handwall.Location2was 11ft8inintotheroomalongaline7ft0.5ininfromtheleft-handwall.Location3was 50cmintotheroomalongaline7ft0.5ininfromtheleft-handwall.Foreachlocation,the 167 KEMARwasfacingawayfromthedoorwhenrotatedtozerodegreesandtheloudspeaker wasfacingthedoor. Inthe\laboratory"twolocationsnearthemiddleoftheroomwereused.Inlocation1 theKEMARwasfacingawayfromthedoorforthezerodegreerotation,andinlocation2 theKEMARwasfacingawayfromthedoor.Thesamefouranglesandthreedistanceswere used. ThereweretwoSAMsignalsproducedbyaTuckerDavisSystemIIwithDD1digital- analogandanalog-digitalconverter.Thesampleratewas50kHzwith16bits/sample.Both signalshada4-kHzcarrierandwere5secondslongwithhardonsetsandThe amplitudemodulationhadratesofeither100Hzor500Hzandhadadepthof100%.A RadioShackMinimus3.5,consistingofa2.5-indriverinasealedboxwasusedtogenerate thetonesatalevelof64dBA.Binauralrecordingsweremadeatthesametimeasthe5- secondpresentationofthetones.ThesignalsfromtheKEMARwerefurther withanAudioBuddypThegainsoftheleftandrightpre-am werecalibratedtobewithin1dBofeachother. SummaryofData Fromeachrecordingthefollowingwerecalculated:theILD,theEITD,theinterauralenve- lopecoherence,theAMineachear, m left and m right ,andtheQFMineachear, left and right .Forthereverberationroom,thesequantitiesarepresentedinTable.1for100Hzand Table.2for500Hz.Forthe\laboratory"thequantitiesarepresentedinTable.3for100Hz andTable.4for500Hz. Comparedtothevaluesof m inFig.2.8for100-HzAMinfreethevaluesof m 168 inTable.1forthereverberationroomshowmorevariation.Thisisalsothecasebuttoa lesserdegreefor m inthe\laboratory"(Table.3).Infree m wasneverfoundtobe muchsmallerthanabout0.75.Inthereverberationroomandthe\laboratory"thereare instanceswithmuchsmallervalueswhichmayhaveimportantperceptualconsequenceson thestrengthoftheEITDcue[40].The500-HzmodulationfrequencydatainTables.2and .4alsocontainconsiderablevariationin m withmorevariationinthereverberationroom. TheQFM( )alsoshowedmorevariationinroomsthaninfree(Fig.2.9).Infree rarelyexceededavalueof1.Inrooms,thereareseveralinstancesofthis,including oneverylargevalueat500Hzinthe\laboratory"of20.6270.Because isverysensitive tochangesinphase,itisnotsurprisingthatroomwouldleadtothistypeof distribution. TheEITDcueisinconsistentwithazimuthalangleinrooms.Changesinthedistance tothesource,andchangesinthelocationintheroomcreateanEITDthatappearstobe mostlyrandom.TheEITDisoftennegative,eventhoughtheazimuthalanglesarelocated between0and90degrees. Theinterauralenvelopecoherenceremainedlargeinrooms.Acomputersimulationof n =1000randompairsofenvelopeswithRayleigh-distributedamplitudes,anduniformly- distributedphasesshowedthattheinterauralenvelopecoherencedoesnotdeviatefromunity bymuch.Themeanvaluewas =0 : 9979,thestandarddeviationwas ˙ =0 : 0012,andthe standarderrorwas ˙ p 1000 =0 : 000039165. TheILDappearedtobeamorereliablecueinroomsthattheEITD.Therewerefewer misleading(negative)cues.TheILDappearedtoincreasewithazimuthalanglemorereliably thantheEITD. 169 location1 location2 location3 d (cm) 50 100 200 50 100 200 50 100 200 0 m left 1.5445 0.8756 0.4745 1.4099 0.4198 1.7936 1.3937 0.5885 0.1531 m right 0.7930 0.6594 0.6875 0.7951 0.4993 0.3572 0.8753 0.3950 0.7184 left 0.4768 0.4580 0.8928 0.5495 0.5423 0.3604 0.0717 0.5822 0.2245 right 0.1213 0.3015 0.2776 0.1236 0.1907 0.9751 0.2598 0.4194 0.8649 EITD 260 -520 1560 -340 1080 -20 -180 -440 2400 E.Coh. 0.9850 0.9971 0.9745 0.9943 0.9932 0.9217 0.9864 0.9938 0.9565 ILD 4.7 0.3 -0.6 3.0 4.3 5.9 2.4 0.6 -1.4 30 m left 0.9720 0.3107 0.2546 0.7301 2.1296 1.8191 0.6172 0.3607 0.4048 m right 1.1057 0.7871 0.6929 0.9049 1.5508 0.3378 0.9741 0.7965 0.9068 left 0.7020 1.2994 0.1239 0.1976 2.5768 1.7730 0.1594 0.0959 0.6591 right 0.0398 0.3262 0.1428 0.0623 0.4187 0.3310 0.1626 0.2097 0.1582 EITD 0 -4000 3320 200 2960 2160 140 -140 -5000 E.Coh. 0.9863 0.9182 0.9549 0.9936 0.9296 0.9594 0.9846 0.9681 0.9307 ILD 15.3 11.6 4.4 10.9 9.6 11.9 9.4 7.9 0.8 60 m left 0.6119 0.2347 1.2575 0.3284 0.8312 0.3034 0.5120 3.8221 1.1431 m right 1.1285 1.0267 0.8148 0.9464 1.8968 0.5433 0.8261 1.0279 0.5721 left 0.1917 0.4958 0.9659 0.1997 0.4159 0.3903 0.0463 3.3003 0.4881 right 0.1511 0.3505 0.5763 0.0322 1.0464 0.5704 0.2553 0.2946 0.4132 EITD 140 460 -2280 580 -20 200 880 -340 1260 E.Coh. 0.9730 0.9208 0.9942 0.9326 0.9586 0.9895 0.9852 0.8871 0.9797 ILD 10.2 9.7 2.6 10.8 2.9 -0.2 9.5 12.4 7.1 90 m left 0.4338 0.2410 2.1282 0.7818 2.0527 0.3172 0.6372 0.3197 0.6171 m right 1.2117 0.4886 0.4477 1.2439 2.2506 0.1616 0.9738 0.9701 0.7463 left 0.1921 0.3995 1.1502 0.3582 0.4298 0.2957 2.5129 0.5764 0.3911 right 0.1828 0.3078 0.9657 0.0525 1.0517 0.4135 0.1975 0.5098 0.5591 EITD 1560 -520 -580 720 1960 -3980 3700 3840 -2100 E.Coh. 0.9402 0.9850 0.9732 0.9800 0.9916 0.9842 0.8470 0.9527 0.9981 ILD 11.0 6.6 9.6 7.0 1.5 -1.9 16.9 -2.1 0.7 Table.1MeasurementsforSAMtoneinthereverberationroomatamodulationfrequency of100Hz.Foreachangle,locationanddistance( d ),quantitiesshownaretheAMinthe leftear( m left ),theAMintherightear( m right ),theQFMintheleftear( left ),theQFM intherightear( right ),theenvelopeinterauraltime(EITD)in s,theinteraural envelopecoherence(E.Coh.),andtheinterauralleveldi(ILD)indB. 170 location1 location2 location3 d (cm) 50 100 200 50 100 200 50 100 200 0 m left 0.8379 0.6355 0.5402 1.5785 0.60422 1.8565 1.0388 0.5980 0.3559 m right 0.8589 0.7111 1.0643 0.5676 0.8413 0.5275 0.5028 0.7415 1.8287 left 0.1425 0.5565 0.2925 0.4589 1.2760 1.0355 0.2725 0.8793 1.2703 right 0.1019 0.5293 0.3770 0.2711 0.3469 1.2142 0.4989 0.6458 0.9776 EITD -20 40 -140 -20 -320 -800 120 -140 -900 E.Coh. 0.9997 0.9931 0.9719 0.9752 0.9587 0.9649 0.9548 0.9928 0.9333 ILD 7.2 1.0 1.4 2.5 2.8 5.8 2.8 1.2 -1.2 30 m left 1.4466 0.9902 0.4199 0.5205 1.0475 0.9736 0.2044 0.2446 0.5554 m right 1.0008 0.8964 0.4651 0.9489 2.1246 0.1898 0.6593 0.5381 0.6729 left 1.5811 0.0486 0.2518 0.3208 3.3604 1.1770 0.1835 0.2292 0.5745 right 0.0415 0.3027 0.3982 0.2645 0.6631 0.5310 0.1594 0.5813 0.6689 EITD 260 300 -100 180 -620 -320 100 20 260 E.Coh. 0.9653 0.9949 0.9994 0.9775 0.9098 0.9575 0.9579 0.9897 0.9941 ILD 12.4 13.8 3.9 11.2 10.9 14.4 9.2 7.7 0.2 60 m left 0.5608 0.9908 0.6755 0.4783 0.5317 0.6571 0.6676 2.0886 1.6293 m right 1.1370 0.8289 0.3514 0.8255 1.1904 0.7945 0.6645 0.8727 0.9892 left 0.6541 0.8601 0.3369 0.3402 0.8234 0.3010 0.1342 1.2437 0.5079 right 0.1720 0.5388 0.7662 0.4163 2.0632 0.7148 0.2146 0.6813 0.5951 EITD 540 720 740 660 80 420 520 -380 880 E.Coh. 0.9465 0.9951 0.9628 0.9852 0.9591 0.9923 0.9987 0.9757 0.9843 ILD 9.7 7.0 3.6 10.2 3.0 0.4 8.7 14.1 7.0 90 m left 0.6543 1.1516 1.2218 0.7902 0.9061 1.0965 3.2058 0.3902 1.2611 m right 1.4227 0.5725 0.1086 0.9213 2.9357 0.6965 0.8293 1.5403 0.9417 left 0.5657 0.7351 3.8372 0.5685 1.5851 1.2577 1.7952 0.5668 0.6452 right 0.3893 0.7280 0.1497 0.5919 1.7889 0.3479 0.5439 0.9004 0.3338 EITD -500 400 -740 280 40 40 -800 560 -20 E.Coh. 0.9734 0.9687 0.9456 0.9907 0.9822 0.9913 0.9772 0.9569 0.9986 ILD 10.7 5.1 3.8 6.2 3.4 -4.5 14.7 0.4 -1.2 Table.2SameasTable.1butforthereverberationroomat500Hz. 171 location1 location2 d (cm) 50 100 200 50 100 200 0 m left 0.8100 0.7885 0.8104 1.0872 0.6181 0.8844 m right 0.9330 0.9409 1.4727 0.9144 0.8108 0.7636 left 0.0156 0.1622 0.5390 0.2282 0.1078 0.6571 right 0.1079 0.3242 0.7879 0.2436 0.2999 0.6011 EITD 180 240 120 -260 340 -400 E.Coh. 0.9979 0.9987 0.9809 0.9966 0.9966 0.9983 ILD 0.6 3.2 0.8 2.0 -1.6 -0.5 30 m left 1.4911 0.7394 1.666 0.8435 1.5139 0.1340 m right 0.9759 1.0049 0.9603 0.8370 1.1159 0.5985 left 0.2945 0.8047 0.9547 0.0475 1.2610 0.4465 right 0.1039 0.0846 0.2474 0.0759 0.0886 0.0963 EITD -120 2560 -1600 760 1820 0 E.Coh. 0.9912 0.9700 0.9895 0.9981 0.9606 0.9514 ILD 13.9 12.6 11.3 11.8 12.2 9.1 60 m left 0.6251 0.8951 0.5914 1.1320 1.2511 0.3153 m right 0.9190 1.0828 1.0762 0.9178 1.1978 0.8949 left 0.3132 0.7596 0.3808 0.5171 0.4671 1.3520 right 0.0836 0.3412 0.0697 0.1208 0.2779 0.4185 EITD 600 -500 220 780 -80 -4020 E.Coh. 0.9809 0.9836 0.9656 0.9900 0.9980 0.9180 ILD 16.3 9.2 5.3 13.8 11.4 17.0 90 m left 0.4067 1.1268 0.9762 0.2972 0.1390 0.8198 m right 0.9998 1.2441 4.0648 0.9250 0.9999 3.3393 left 0.4139 0.8395 1.6093 1.6407 0.1923 0.3819 right 0.1164 0.3821 7.7200 0.2586 0.2372 1.7341 EITD 2920 1640 2820 -1060 1340 2820 E.Coh. 0.9364 0.9872 0.9360 0.8775 0.8851 0.9177 ILD 19.2 9.3 0.6 14.9 5.1 -2.4 Table.3SameisTable.1butforthethe\laboratory"at100Hz. 172 location1 location2 d (cm) 50 100 200 50 100 200 0 m left 0.9875 1.4277 1.3406 1.1955 0.7131 0.4907 m right 0.9491 1.0009 1.2350 0.9099 0.9230 0.4349 left 0.2922 0.1002 0.7831 0.2526 0.1540 1.3852 right 0.2477 0.4749 0.7430 0.3113 0.4237 1.2396 EITD 0 40 80 -40 60 -60 E.Coh. 0.9996 0.9906 0.9891 0.9946 0.9973 0.9847 ILD 0.3 1.4 -1.9 1.8 -1.4 -0.3 30 m left 1.2242 0.3694 1.8766 0.7852 1.0914 0.5982 m right 1.0469 0.9813 0.6639 0.9128 1.0379 0.4634 left 0.7052 1.4314 1.7679 0.2565 1.1859 0.8868 right 0.1375 0.1325 0.4823 0.0973 0.4175 0.3340 EITD 160 -540 560 140 120 840 E.Coh. 0.9959 0.8899 0.9902 0.9968 0.9752 0.9907 ILD 14.4 11.9 9.5 12.2 13.8 7.9 60 m left 1.0428 1.2232 0.4876 0.7511 0.6937 2.9010 m right 1.0330 0.9984 1.4846 1.0274 1.1181 0.4452 left 0.3208 0.6330 0.5706 0.7551 1.4402 1.3217 right 0.2377 0.5541 0.6594 0.3272 1.0080 0.2584 EITD 560 580 660 580 480 -480 E.Coh. 0.9981 0.9988 0.9532 0.9778 0.9621 0.9492 ILD 15.3 9.0 5.3 15.1 11.2 11.8 90 m left 1.2183 1.364 1.4309 1.0319 0.8004 1.0140 m right 0.9477 1.0277 6.2806 0.9665 1.3526 5.0144 left 0.3307 1.6183 1.1942 0.4966 0.0670 1.2269 right 0.3815 1.0628 20.6270 0.5411 0.9354 2.7867 EITD 740 800 460 940 -40 -840 E.Coh. 0.9925 0.9781 0.9025 0.9931 0.9906 0.9464 ILD 17.6 7.1 7.0 16.9 6.0 -0.3 Table.4SameisTable.1butforthethe\laboratory"at500Hz. 173 BIBLIOGRAPHY 174 BIBLIOGRAPHY [1]G.Andeol,E.A.Macpherson,andA.T.Sabin.Soundlocalizationinnoiseandsensi- tivitytospectralshape. Hear.Res. ,304:20{27,2013. [2]L.R.BernsteinandC.Trahiotis.Lateralizationoflow-frequency,complexwaveforms: Theuseofenvelope-basedtemporaldisparities. J.Acoust.Soc.Am. ,77(5),1985. [3]L.R.BernsteinandC.Trahiotis.Lateralizationofsinusoidallyamplitude-modulated tones:ofspectrallocusandtemporalvariation. J.Acoust.Soc.Am. ,78(2):514{ 523,1985. [4]L.R.BernsteinandC.Trahiotis.Enhancingsensitivitytointerauraldelaysathigh frequenciesbyusing\transposedstimuli". J.Acoust.Soc.Am. ,112(3):1026{1036,2002. [5]L.R.BernsteinandC.Trahiotis.Howsensitivitytoongoinginterauraltemporaldispar- itiesisbymanipulationsoftemporalfeaturesoftheenvelopesofhigh-frequency stimuli. J.Acoust.Soc.Am. ,125(5):3234{3242,2009. [6]J.Blauert.Soundlocalizationinthemedianplane. Acustica ,22:205{213,1969-70. [7]A.Brughera,L.Dunai,andW.M.Hartmann.Humaninterauraltimethresh- oldsforsinetones:Thehigh-frequencylimit. J.Acoust.Soc.Am. ,133(5),2013. [8]M.D.BurkhardandR.M.Sachs.Anthropometricmanikinforacousticresearch. J. Acoust.Soc.Am. ,58(1),1975. [9]R.A.Butler,S.K.andR.F.Naunton.Theroleofstimulusfrequencyinthe localizationofsoundinspace. J.Aud.Res. ,7:169{180,1967. [10]M.Dietz,L.R.Bernstein,C.Trahiotis,S.D.Ewert,andV.Hohmann.Theof overalllevelonsensitivitytointerauraloftimeandlevelathighfrequencies. J.Acoust.Soc.Am. ,134(1):494{502,2013. [11]G.Eberle,K.I.McAnally,R.L.Martin,andP.Flanagan.Localizationofamplitude- modulatedhigh-frequencynoise. J.Acoust.Soc.Am. ,107(6),2000. 175 [12]R.Fisher. StatisticalMethodsForResearchWorkers .Cosmostudyguides.Cosmo Publications,1925. [13]B.R.GlasbergandB.C.Moore.Derivationofauditoryshapesfromnotched-noise data. Hear.Res. ,47(1):103{138,1990. [14]M.J.GoupellandW.M.Hartmann.Interauralandthedetectionof interauralincoherence.III.Narrowbandexperimentsandbinauralmodels. J.Acoust. Soc.Am. ,122(2),2007. [15]R.V.L.HartleyandT.C.Fry.Thebinaurallocationofpuretones. Phys.Rev. , 18:431{442,Dec1921. [16]W.M.Hartmann. Signals,Sound,andSensation .ModernAcousticsandSignalPro- cessing.Springer-Verlag,NewYork(AmericanInstituteofPhysicsPress),1998. [17]W.M.HartmannandE.J.Macaulay.Anatomicallimitsoninterauraltime Anecologicalperspective. FrontiersinNeuroscience ,8(34),2014. [18]W.M.HartmannandB.Rakerd.LocalizationofsoundinroomsIV:TheFranssen J.Acoust.Soc.Am. ,86(4),1989. [19]W.M.Hartmann,B.Rakerd,andA.Koller.Binauralcoherenceinrooms. Actaacustica unitedwithacustica ,91(3):451{462,2005. [20]W.M.HartmannandA.Wittenberg.Ontheexternalizationofsoundimages. J. Acoust.Soc.Am. ,99(6),1996. [21]W.M.HartmannandE.M.Wolf.Matchingthewaveformandthetemporalwindow inthecreationofexperimentalsignals. J.Acoust.Soc.Am. ,126(5),2009. [22]E.Hecht.Optics.PearsonEducation,page496.Addison-Wesley,4thedition,2002. [23]G.B.Henning.Detectabilityofinterauraldelayinhigh-frequencycomplexwaveforms. J.Acoust.Soc.Am. ,55(1),1974. [24]G.B.Henning.Someobservationsonthelateralizationofcomplexwaveforms. J. Acoust.Soc.Am. ,68(2),1980. [25]G.B.Henning.Lateralizationoflow-frequencytransients. Hear.Res. ,9(2):153{172, 1983. 176 [26]G.B.HenningandJ.Ashton.Theofcarrierandmodulationfrequencyon lateralizationbasedoninterauralphaseandinterauralgroupdelay. Hear.Res. ,4(2):185 {194,1981. [27]G.F.Kuhn.Modelfortheinterauraltimeintheazimuthalplane. J.Acoust. Soc.Am. ,62(1),1977. [28]G.F.Kuhn.Thepressuretransformationfromadisoundldtotheexternalear andtothebodyandheadsurface. J.Acoust.Soc.Am. ,65(4),1979. [29]G.F.Kuhn.Physicalacousticsandmeasurementspertainingtodirectionalhearing.In W.YostandG.Gourevitch,editors, DirectionalHearing ,Proceedingsinlifesciences, pages3{25.Springer-Verlag,1987. [30]D.M.Leakey,B.M.Sayers,andC.Cherry.Binauralfusionoflow-andhigh-frequency sounds. J.Acoust.Soc.Am. ,30(3):222{222,1958. [31]E.J.Macaulay,W.M.Hartmann,andB.Rakerd.Theacousticalbrightspotand mislocalizationoftonesbyhumanlisteners. J.Acoust.Soc.Am. ,127(3),2010. [32]E.A.MacphersonandJ.C.Middlebrooks.Listenerweightingofcuesforlateralangle: Theduplextheoryofsoundlocalizationrevisited. J.Acoust.Soc.Am. ,111(5),2002. [33]D.McFaddenandE.G.Pasanen.Lateralizationathighfrequenciesbasedoninteraural time J.Acoust.Soc.Am. ,59(3),1976. [34]A.W.Mills.Ontheminimumaudibleangle. J.Acoust.Soc.Am. ,30(4),1958. [35]A.W.Mills.Auditorylocalization.InJ.Tobias,editor, FoundationsofModernAuditory Theory ,volume2,pages301{345.AcademicPress,1970. [36]S.M.MisurelliandR.Y.Litovsky.Spatialreleasefrommaskinginchildrenwithnormal hearingandwithbilateralcochlearimplants:ofinterfererasymmetry. J.Acoust. Soc.Am. ,132(1),2012. [37]J.J.M.Monaghan,K.Krumbholz,andB.U.Seeber.Factorstingtheuseof envelopeinterauraltimeinreverberation. J.Acoust.Soc.Am. ,133(4),2013. [38]P.M.C.Morse. Vibrationandsound .Internationalseriesinphysics.McGraw-Hillbook company,inc.,1936. 177 [39]J.O.Nordmark.Binauraltimediscrimination. J.Acoust.Soc.Am. ,60(4),1976. [40]J.M.NuetzelandE.R.Hafter.Discriminationofinterauraldelaysincomplexwave- forms:Spectralects. J.Acoust.Soc.Am. ,69(4),1981. [41]M.Otani,T.Hirahara,andS.Ise.Numericalstudyonsource-distancedependencyof head-relatedtransferfunctions. J.Acoust.Soc.Am. ,125(5),2009. [42]B.RakerdandW.M.Hartmann.Localizationofsoundinrooms,III:Onsetand duration J.Acoust.Soc.Am. ,80(6):1695{1706,1986. [43]B.RakerdandW.M.Hartmann.Localizationofsoundinrooms.V.Binauralcoherence andhumansensitivitytointerauraltimeinnoise. J.Acoust.Soc.Am. , 128(5),2010. [44]S.N.Rschevkin. ACourseofLecturesontheTheoryofSound .PergamonnPress,1963. [45]B.G.Shinn-Cunningham,S.Santarelli,andN.Kopco.Toriofconfusion:Binaural localizationcuesforsourceswithinreachofalistener. J.Acoust.Soc.Am. ,107(3):1627{ 1636,2000. [46]M.A.Stellmack,N.F.Viemeister,andA.J.Byrne.Discriminationofinteraural phaseintheenvelopesofsinusoidallyamplitude-modulated4-khztonesasa functionofmodulationdepth. J.Acoust.Soc.Am. ,118(1),2005. [47]J.W.Strutt. TheTheoryofSoundV.2 .Macmillan,1896. [48]J.W.Strutt.XII.Onourperceptionofsounddirection. PhilosophicalMagazineSeries 6 ,13(74):214{232,1907. [49]F.L.WightmanandD.J.Kistler.Thedominantroleoflow-frequencyinterauraltime insoundlocalization. J.Acoust.Soc.Am. ,91(3),1992. [50]W.A.Yost.Lateralpositionofsinusoidspresentedwithinterauralintensiveandtem- porald J.Acoust.Soc.Am. ,70(2),1981. [51]P.X.ZhangandW.M.Hartmann.Ontheabilityofhumanlistenerstodistinguish betweenfrontandback. Hear.Res. ,260(12):30{46,2010. [52]J.ZwislockiandR.S.Feldman.Justnoticeableindichoticphase. J.Acoust. Soc.Am. ,28(5),1956. 178