心理语言学：CH3.perception - 副本

Chapter3:SpeechPerception,OverviewofQuestions,Cancomputersperceivespeechaswellashumans?Whydoesanunfamiliarforeignlanguageoftensoundlikeacontinuousstreamofsound,withnobreaksbetweenwords?Doeseachwordthatwehearhaveauniquepatternofairpressurechangesassociatedwithit?Aretherespecificareasinthebrainthatareresponsibleforperceivingspeech?,Speechperceptionreferstotheprocessesbywhichhumansareabletointerpretandunderstandthesoundsusedinlanguage.Thestudyofspeechperceptioniscloselylinkedtothefieldsofphoneticsandphonologyinlinguisticsandcognitivepsychologyandperceptioninpsychology.,Researchinspeechperceptionseekstounderstandhowhumanlistenersrecognizespeechsoundsandusethisinformationtounderstandspokenlanguage.Speechresearchhasapplicationsinbuildingcomputersystemsthatcanrecognizespeech,aswellasimprovingspeechrecognitionforhearing-andlanguage-impairedlisteners.,SpeechPerception,Thefirststepincomprehendingspokenlanguageistoidentifythewordsbeingspoken,performedinmultiplestages:1.Phonemesaredetected(/b/,/e/,/t/,/e/,/r/,)2.Phonemesarecombinedintosyllables(/be/ter/)3.Syllablesarecombinedintowords(“better”)4.Wordmeaningretrievedfrommemory,Spectrogram:Ioweyouayo-yo,Speechperception:twoproblems,Wordsarenotneatlysegmented(e.g.,bypauses)LackofphonemeinvarianceCoarticulation=consecutivespeechsoundsblendintoeachotherduetomechanicalconstraintsonarticulatorsSpeakerdifferences;pitchaffectedbyageandsex;differentdialects,talkingspeedsetc.,Thespeechinputconsistsof;,Frequencyrange50-5600HzCriticalbandfilters(临界频带滤波器）Dynamicrange50dBTemporalresolution（瞬时清晰度）of10msSmallestdetectablechangeinF02HzSmallestchangeinF140HzSmallestchangeinF2100HzSmallestchangeinF3150Hz,TheSpeechStimulus,Phoneme-smallestunitofspeechthatchangesmeaninginawordInEnglishthereare47phonemes:23majorvowelsounds24majorconsonantsoundsNumberofphonemesinotherlanguagesvaried11inHawaiianand60insomeAfricandialects,Table13.1MajorconsonantsandvowelsofEnglishandtheirphoneticsymbols,TheAcousticSignal,ProducedbyairthatispushedupfromthelungsthroughthevocalcordsandintothevocaltractVowelsareproducedbyvibrationofthevocalcordsandchangesintheshapeofthevocaltractThesechangesinshapecausechangesintheresonantfrequencyandproducepeaksinpressureatanumberoffrequenciescalledformants（共振峰）,Figure13.1Thevocaltractincludesthenasalandoralcavitiesandthepharynx,aswellascomponentsthatmove,suchasthetongue,lips,andvocalcords.,TheAcousticSignal-continued,Thefirstformanthasthelowestfrequency,thesecondhasthenexthighest,etc.SoundspectrogramsshowthechangesinfrequencyandintensityforspeechConsonantsareproducedbyaconstrictionofthevocaltractFormanttransitions-rapidchangesinfrequencyprecedingorfollowingconsonants,Figure13.3Spectrogramofthewordhadshowingthefirst(F1),second(F2),andthird(F3)formantsforthevowel/ae/.(SpectrogramcourtesyofKerryGreen.),Figure13.4Spectrogramofthesentence“Royreadthewill,”showingtheformantssuchasF1,F2,andF3,andformanttransitionssuchasT2andT3.(SpectrogramcourtesyofKerryGreen.),TheRelationshipbetweentheSpeechStimulusandSpeechPerception,Thesegmentationproblem-therearenophysicalbreaksinthecontinuousacousticsignalHowdowesegmenttheindividualwords?Thevariabilityproblem-thereisnosimplecorrespondencebetweentheacousticsignalandindividualphonemesVariabilityfromaphonemescontextCoarticulation-overlapbetweenarticulationofneighboringphonemes,Figure13.5Spectrogramof“Ioweyouayo-yo.”Thisspectrogramdoesnotcontainpausesorbreaksthatcorrespondtothewordsthatwehear.Theabsenceofbreaksintheacousticsignalcreatesthesegmentationproblem.(SpectrogramcourtesyofDavidPisoni.),Figure13.6Hand-drawnspectrogramsfor/di/and/du/.(From“PerceptionoftheSpeechCode,”byA.M.Liberman,1967,PsychologicalReview,74,431-461,figure1.Copyright1967bytheAmericanPsychologicalAssociation.Reprintedbypermissionoftheauthor.),TheRelationshipbetweentheSpeechStimulusandSpeechPerception-continued,VariabilityfromdifferentspeakersSpeakersdifferinpitch,accent,speedinspeaking,andpronunciationThisacousticsignalmustbetransformedintofamiliarwordsPeopleperceivespeecheasilyinspiteofthesegmentationandvariabilityproblems,Figure13.7(a)Spectrogramof“Whatareyoudoing?”pronouncedslowlyanddistinctly.(b)Spectrogramof“Whatareyoudoing?”aspronouncedinconversationalspeech.(SpectrogramcourtesyofDavidPisoni.),StimulusDimensionsofSpeechPerception,Invariantacousticcues-featuresofphonemesthatremainconstantShort-termspectrogramsareusedtoinvestigateinvariantacousticcuesSequenceofshort-termspectracanbecombinedtocreatearunningspectraldisplayFromthesedisplays,therehavebeensomeinvariantcuesdiscovered,Figure13.8Left:ashort-termspectrumoftheacousticenergyinthefirst26msofthephoneme/ga/.Right:soundspectrogramofthesamephoneme.Thesoundforthefirst26msisindicatedinred.Thepeakintheshort-termspectrum,markeda,correspondstothedarkbandofenergy,markedainthespectrum.Theminimumintheshort-termspectrum,markedb,correspondstothelightarea,markedbinthespectrogram.Thespectrogramontherightshowstheenergyfortheentire500msdurationofthesound,whereastheshort-termspectrumonlyshowsthefirst26msatthebeginningofthissignal.(CourtesyofJamesSawusch.),Figure13.9Runningspectraldisplaysfor/pi/and/da/.Thesedisplaysaremadeupofasequenceofshort-termspectra,liketheoneinFigure13.8.Eachofthesespectraisdisplaced5msonthetimeaxis,sothateachstepwemovealongthisaxisindicatesthefrequenciespresentinthenext5ms.Thelow-frequencypeak(V)inthe/da/displayisacueforvoicing.(From“Time-VaryingFeaturesofInitialStopConsonantsinAuditoryRunningSpectra:AFirstReport,”byD.Kewley-Port,andP.A.Luce,1984,PerceptionandPsychophysics,35,353-360,figure1.Copyright1984byPsychonomicSocietyPublications.Reprintedbypermission.),CategoricalPerception,ThisoccurswhenawiderangeofacousticcuesresultsintheperceptionofalimitednumberofsoundcategoriesAnexampleofthiscomesfromexperimentsonvoiceonsettime(VOT)-timedelaybetweenwhenasoundstartsandwhenvoicingbeginsStimuliareda(VOTof17ms)andta(VOTof91ms),CategoricalPerception-continued,ComputerswereusedtocreatestimuliwitharangeofVOTsfromlongtoshortListenersdonotheartheincrementalchanges,insteadtheyhearasuddenchangefrom/da/to/ta/atthephoneticboundaryThus,weexperienceperceptualconstancyforthephonemeswithinagivenrangeofVOT,Figure13.10Spectrogramsfor/da/and/ta/.Thevoiceonsettime-thetimebetweenthebeginningofthesoundandtheonsetofvoicing-isindicatedatthebeginningofthespectrogramforeachsound(SpectrogramcourtesyofRonCole.),Figure13.11Theresultsofacategoricalperceptionexperimentindicatethat/da/isperceivedforVOTstotheleftofthephoneticboundary,andthat/ta/isperceivedatVOTstotherightofthephoneticboundary.(From“SelectiveAdaptationofLinguisticFeatureDetectors,byP.EimasandJ.D.Corbit,1973,CognitivePsychology,4,99-109,figure2.Copyright1973AcademicPress,Inc.Reprintedbypermission.),Figure13.12Inthediscriminationpartofacategoricalperceptionexperiment,twostimuliarepresented,andthelistenerindicateswhethertheyarethesameordifferent.ThetypicalresultisthattwostimuliwithVOTsonthesamesideofthephoneticboundary(solidarrows)arejudgedtobethesame,andthattwostimuliondifferentsidesofthephoneticboundary(dashedarrows)arejudgedtobedifferent.,Figure13.13PerceptualconstancyoccurswhenallstimuliononesideofthephoneticboundaryareperceivedtobeinthesamecategoryeventhoughtheirVOTischangedoverasubstantialrange.ThisdiagramsymbolizestheconstancyobservedbyEimasandCorbit(1973)experiment,inwhich/da/washeardononesideoftheboundaryand/ta/ontheotherside.,SpeechPerceptionisMultimodal,Auditory-visualspeechperceptionTheMcGurkeffectVisualstimulusshowsaspeakersaying“ga-ga”Auditorystimulushasaspeakersaying“ba-ba”Observerwatchingandlisteninghears“da-da”,whichisthemidpointbetween“ga”and“ba”Observerwitheyesclosedwillhear“ba”,McGurkEffect,Figure13.14TheMcGurkeffect.Thewomanslipsaremovingasifsheissaying/ga-ga/,buttheactualsoundbeingpresentedis/ba-ba/.Thelistener,however,reportshearingthesound/da-da/.Ifthelistenercloseshiseyes,sothathenolongerseesthewomanslips,hehears/ba-ba/.Thus,seeingthelipsmovinginfluenceswhatthelistenerhears.,CognitiveDimensionsofSpeechPerception,Top-downprocessing,includingknowledgealistenerhasaboutalanguage,affectsperceptionoftheincomingspeechstimulusSegmentationisaffectedbycontextandmeaningIscreamyouscreamweallscreamforicecream,Figure13.15Speechperceptionistheresultoftop-downprocessing(basedonknowledgeandmeaning)andbottom-upprocessing(basedontheacousticsignal)workingtogether.,MeaningandPhonemePerception,ExperimentbyTurveyandVanGelderShortwords(sin,bat,andleg)andshortnonwords(jum,baf,andteg)werepresentedtolistenersThetaskwastopressabuttonasquicklyaspossiblewhentheyheardatargetphonemeOnaverage,listenerswerefasterwithwords(580ms)thannon-words(631ms),MeaningandPhonemePerception-continued,ExperimentbyWarrenListenersheardasentencethathadaphonemecoveredbyacoughThetaskwastostatewhereinthesentencethecoughoccurredListenerscouldnotcorrectlyidentifythepositionandtheyalsodidnotnoticethataphonemewasmissing-calledthephonemicrestorationeffect,Phonemicrestoration,AuditorypresentationPerceptionLegislaturelegislatureLegi_laturelegilatureLegi*laturelegislatureItwasfoundthatthe*eelwasontheaxle.wheelItwasfoundthatthe*eelwasontheshoe.heelItwasfoundthatthe*eelwasontheorange.peelItwasfoundthatthe*eelwasonthetable.meal,Warren,R.M.(1970).Perceptualrestorationsofmissingspeechsounds.Science,167,392-393.,MeaningandWordPerception,ExperimentbyMillerandIsardStimuliwerethreetypesofsentences:NormalgrammaticalsentencesAnomaloussentencesthatweregrammaticalUngrammaticalstringsofwordsListenersweretoshadow(repeataloud)thesentencesastheyheardthemthroughheadphones,MeaningandWordPerception-continued,Resultsshowedthatlistenerswere89%accuratewithnormalsentences79%accurateforanomaloussentences56%accurateforungrammaticalwordstringsDifferenceswereevenlargerifbackgroundnoisewaspresent,SpeakerCharacteristics,Indexicalcharacteristics-characteristicsofthespeakersvoicesuchasage,gender,emotionalstate,levelofseriousness,etc.ExperimentbyPalmeri,Goldinger,andPisoniListenersweretoindicatewhenawordwasnewinasequenceofwordsResultsshowedthattheyweremuchfasterifthesamespeakerwasusedforallthewords,SpeechPerceptionandtheBrain,Brocasaphasia-individualshavedamageinBrocasarea(infrontallobe)LaboredandstiltedspeechandshortsentencesbuttheyunderstandothersWernickesaphasia-individualshavedamageinWernickesarea(intemporallobe)SpeakfluentlybutthecontentisdisorganizedandnotmeaningfulTheyalsohavedifficultyunderstandingothers,Figure13.16BrocasandWernickesareas,whicharespecializedforlanguageproductionandcomprehension,arelocatedinthelefthemisphereofthebraininmostpeople.,SpeechPerceptionandtheBrain-continued,MeasurementsfromcatsauditoryfibersshowthatthepatternoffiringmirrorstheenergydistributionintheauditorysignalBrainscansofhumansshowthatthereareareasofthehumanwhatstreamthatareselectivelyactivatedbythehumanvoice,Figure13.17(a)Short-termspectrumfor/da/.Thiscurveindicatestheenergydistributionin/da/between20and40msafterthebeginningofthesignal.(b)Nervefiringofapopulationofcatauditorynervefiberstothesamestimulus.(From“EncodingofSpeechFeaturesintheAuditoryNerve,”byM.B.Sachs,E.D.Young,andM.I.Miller,1981.InR.CarlsonandB.Granstrom(Eds.)TheRepresentationofSpeechinthePeripheralAuditorySystem,pp.115-130.Copyright1981byElsevierSciencePublishing,NewYork.Reprintedbypermission.),ExperienceDependentPlasticity,Beforeage1,humaninfantscantelldifferencebetweensoundsthatcreatealllanguagesThebrainbecomes“tuned”torespondbesttospeechsoundsthatareintheenvironmentOthersounddifferentiationdisappearswhenthereisnoreinforcementfromtheenvironment,MotorTheoryofSpeechPerception,LposedthatmotormechanismsresponsibleforproducingsoundsactivatemechanismsforperceivingsoundEvidencefrommonkeyscomesfromtheexistenceofmirrorneuronsExperimentbyWatkinsetal.Participantshadtheirmotorcortexforfacemovementsstimulatedbytranscranialmagneticstimulation(TMS),MotorTheoryofSpeechPerception-continued,Resultsshowedsmallmovementsforthemouthcalledmotorevokedpotentials(MEP)ThisresponsebecamelargerwhenthepersonlistenedtospeechorwatchedsomeoneelseslipmovementsInaddition,thewherestreammayworkwiththewhatstreamforspeechperception,Figure13.18Thetranscranialmagneticstimulationexperimentthatprovidesevidenceforalinkbetweenspeechperceptionandproductioninhumans.Seetextfordetails.(R