Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
FinalversionacceptedforpublicationinLanguageandSpeechJune6,2017.
Laylistenerclassificationandevaluationoftypicalandatypical
children’sspeech
MelissaA.Redford1,VsevolodKapatsinski1,JolynnCornell-Fabiano2
1DepartmentofLinguistics,UniversityofOregon
2CésarE.ChávezElementarySchool,EugeneSchoolDistrict4J
Runninghead:Classificationandevaluationofchildspeech
Keywords:Autism,likeability,intelligibility,prosody,acoustics.
Acknowledgments.ThisresearchwasfundedbyAwardNumberR01HD061458(PI:
Redford)fromtheEuniceKennedyShriverNationalInstituteofChildHealth&Human
Development(NICHD),andmadepossible,inpart,bythesupportoftheEugeneSchool
District4Jandafellowshiptothe1stauthorfromtheEuropeanInstitutesforAdvanced
Study(EURIAS),co-fundedbytheEuropeanCommission(Marie-Sklodowska-CurieActions
COFUNDProgrammeFP7).Thecontentissolelytheresponsibilityoftheauthorsanddoes
notnecessarilyreflecttheviewsofNICHD,theEugeneSchoolDistrict4J,EURIASor
COFUND.WearegratefultoAubrianneCarson,WookKyungChoe,andPaulOlejarczukfor
significanthelpwithdatacollectionandprocessing.
Classification and evaluation of child speech
1
ABSTRACT
Verbalchildrenwithautismspectrumdisorder(ASD)oftenalsohaveatypicalspeech.In
thecontextofthemanychallengesassociatedwithASD,dospeechsoundpattern
differencesreallymatter?Thecurrentstudyaddressedthisquestion.Structured
spontaneousspeechwaselicitedfrom34children:17withASD,whosecliniciansreported
unusualspeechprosody;and17typically-developing,age-matchedcontrols.Multiword
utteranceswereexcerptedfromeachchild’sspeechsampleandpresentedtoyoungadult
listeners,whohadnoclinicaltrainingorexperience.InExperiment1,listenersclassified
bandpassfilteredandunalteredexcerptsas“typical”or“disordered”.ChildrenwithASD
wereonlydistinguishedfromtypicalchildrenbasedonunalteredspeech,buttheanalyses
indicateduniquecontributionsfromspeechsoundpatterns.InExperiment2,listeners
providedlikeabilityratingsonthefilteredandunalteredexcerpts.Again,laylistenersonly
distinguishedchildrenwithASDfromtheirtypicallydevelopingpeersbasedonunaltered
speech,withtypicalchildrenratedasmorelikeablethanchildrenwithASD.InExperiment
3,listenersevaluatedtheunalteredspeechalongseveralperceptualdimensions.High
overlapbetweenthedimensionsofarticulation,clearness,andfluencywascapturedbyan
emergentdimension:intelligibility.Thisdimensionpredictedlisteners’likeabilityratings
nearlyaswellasitpredictedtheirjudgmentsofdisorder.Overall,theresultsshowthatlay
listenerscandistinguishatypicalfromtypicalchildrenoutsidethesocial-interactional
contextbasedsolelyonspeech,andthattheyattendtospeechintelligibilitytodothis.Poor
intelligibilityalsocontributestolisteners’negativesocialevaluationofchildren,andso
meritsassessmentandremediation.
Classification and evaluation of child speech
2
INTRODUCTION
TempleGrandin,diagnosedwithautismasachild,characterizedherearlyspeechin
thefollowingway:"Myvoicewasflatwithlittleinflectionandnorhythm.Thatalone
stampedmeasdifferent(Grandin&Scariano,1986:21).”Especiallyinchildhood,any
differencethatdeviatesfrompeer-definednormscanjeopardizeachild’saccesstopeer
friendshipsandsocialsupport.Forexample,inthedomainofspeechandlanguage,Rice
andcolleagueshaveshownthatchildrenwithspecificlanguageimpairmentandthosewho
learnEnglishasasecondlanguagehavelessaccesstopositivesocialinteractionsinUS
schoolsthantypicallydeveloping,nativeEnglish-speakingchildren(Rice,Sell,&Hadley,
1991;Gertner,Rice,&Hadley,1994).Atypicalspeechpatternsmayfunctionlikelanguage
disorderorincompletesecondlanguageacquisitionbycreatingbarrierstosocial
interactionforchildrenwithASD.Butunlikelanguagedisorderorincompletesecond
languageacquisition,theproblemmayhavelesstodowithcommunicationperseand
everythingtodowiththeconveyanceofdifference;specifically,adifferencethatresultsin
anegativesocialevaluation.ThispossibilitywasnotedevenbyAsperger(1944,as
translatedbyFrith,1991:70),whoconcludedhisdescriptionofatypicalspeechinchildren
withASDbysayingthat“thelanguagefeelsunnatural,oftenlikeacaricature,which
provokesridiculeinthenaïvelistener.”Thegoalofthepresentresearchwastotestthe
hypothesizedlinkbetweenatypicalspeechpatternsandanegativesocialevaluationof
childrenwithASD.Asecondarygoalwastoidentifythepatternsthatlaylistenersattendto
whenclassifyingchildren’sspeechastypicalornotandwhenevaluatingchildren’s
likeabilityasafunctionoftheirspeech.
Classification and evaluation of child speech
3
Background
Afairlysubstantialbodyofscientificworkconfirmsthatindividualswithautism
whospeakdoindeedproduceatypicalspeechsoundpatterns,andespeciallyatypical
speechprosody.Forexample,Shribergandcolleagues(Shriberg,Paul,McSweeny,Klin,
Cohen,&Volkmar,2001)usedtheversatileandcomprehensivebutunstandardized
Prosody-VoiceScreeningProfile(PVSP;Shriberg,Kwiatkowski,&Rasmussen,1990)in
theirstudyofspontaneousspeechin30individualswithASD:trainedlistenersevaluated
speechacrossavarietyofpredeterminedprosodicandvoicecategories,including
loudness,pitch,voicequality,resonancequality,phrasing,rate,andstress.Themost
consistentvoicedeficitwasinresonancequality:aninappropriatenasalqualitywasmore
oftenidentifiedinspeechproducedbyindividualswithASDthanbythecontrols.Themost
consistentprosodicdeficitwasstress;acategorythatincludedexcessive/equalstress,
prolongationandblockingtypedisfluencies,aswellasincorrectproductionoflexicalstress
patternsandaccentualpatternsatthelevelofthephrase(labeled“contrastivestress”).The
authorsreportthattheperceiveddifferencesbetweenindividualswithASDandcontrol
speakerswasduetoagreaternumberofperceiveddisfluenciesand“misplacedword
stress.”(p.1107).Insofarasanumberofotherperceptualstudiessimilarlyreportautism-
relateddeficitsintheproductionoflexicalstressandphrasalaccenting(e.g.,Baltaxe&
Guthrie,1987;McCaleb&Prizant,1985;Paul,Augustyn,Klin,&Volkmar,2005;Grossman,
Bemis,Plesa-Skwerer,&Tager-Flusberg,2010),itislikelythatdeficitsin“stress”
productionisadefiningfeatureofatypicalprosodyinASD.
Incontrasttoperceptualstudies,acousticstudiesonautisticspeechhavefocusedon
globalpatternsrelevanttocharacterizingtheprosodicdifferencesinindividualswithASD.
Forexample,Diehletal.(2009)extractedaverageF0(theacousticcorrelateofpitch)and
Classification and evaluation of child speech
4
itsstandarddeviation(=pitchvariability)overlargetemporalwindows(250+
milliseconds)innarrativespeechproducedbyatotalof38childrenandadolescentswith
ASDand38typicalcontrols.Theyfoundthatpitchvariabilitywasgenerallyhigherin
speechproducedbychildrenwithASDthanintypicallydevelopingchildren.Otherstudies
reportsimilarresults(Sharda,Subhardra,Sahay,…&Singh,2010;Bonneh,Levanon,Dean-
Pardon,Lossos,&Adini2011;Depape,Chen,Hall,&Trainor,2012;Nadig&Shaw,2012),
whichcouldbeconsistentwiththeabove-mentionedfindingofimpairedphrasalaccenting
inautisticspeechifpitchvariabilitycapturessomethingaboutaccentualpatterning.
Inadditiontodeficitsinlexicalstressproductionandphrasalaccenting,thereis
someevidencethatindividualswithASDproducespeechatslowerratesthantypical
controls(forareviewseeShriberg,Paul,Black,&vanSanten,2011).Forexample,Diehl
andPaul(2011)usedthesameglobalF0measuresasDiehletal.(2009)onutterances
obtainedfrom24childrenwithASDand22typicallydevelopingchildrenindifferent
subtestsoftheformallystructuredbutunstandardizedProfilingElementsofProsodyin
Speech-Communicationtest(Peppé&McCann,2003),butfoundnodifferencebetweenthe
groups.TheyalsotookmeasuresofF0rangeaswellasutterance-levelmeasuresof
acousticdurationandintensity.Onlydurationdifferencesweresignificant:childrenwith
ASDproducedlongerutterancesonaveragethanthetypicalcontrols.
Studiesthatuseperceptualevaluationormeasureglobalacousticstoidentify
prosodicdeficits/differencesassumeanexpandeddefinitionofprosodyrelativetothat
whichisassumedincarefullinguisticphoneticstudies.Butwhereasthelinguist’sinterest
isintherealizationofabstractmetricalandintonationalstructuresandthealignment
betweenthem,theclinician’sinterestisinidentifyingthosepatternsthatcouldadversely
affectanindividual’sabilitytocommunicatewith—inordertobeacceptedby—peers.The
Classification and evaluation of child speech
5
searchforprosodicdeficitsisthusexpandedtoincludeanysuprasegmentalpatternin
autisticspeechthatmightreliablysignaldifferenceordisorder.
Yet,evenwithitswidepurview,existingresearchonthedifferencesbetween
autisticandneurotypicalspeechreturnsfewconsistentfindings.Considerforexamplethe
contradictoryresultsonthestatusofatypicalpitchvariabilityinindividualswithASD:
Diehlandcolleagues(2009)findsignificantdifferencesbetweenchildrenwithandwithout
ASDinonetask,butDiehlandPaul(2011)findnosuchdifferencesinanothertask.Such
inconsistentresultsmaybeduetotheabsenceofdeficitsinasignificantproportionof
high-functioningindividualswithASD.Althoughsystematicdataonthedistributionof
atypicalprosodyinautismdonotyetexist,ithasbeensuggestedthatprosodicdeficitsmay
onlybeevidentinthespeechofhalfofallindividualswithASD(e.g.,Paul,Shribergetal.,
2005).
Despitecontradictoryfindingsandthepossibilitythatprosodicdisorderoccursin
onlyhalfofallhigh-functioningindividualswithASD,thetopicofprosodic
difference/deficitsinautismhasreceivedincreasingattentioninrecentyears.Whereas
McCannandPeppé(2003)foundthatbetween1980and2002only16studiesonprosody
inautism,ascholar.google.frsearchinearly2017on“prosodyautismprosodicASD”
returned1,010results.Thetop100results,whichincludedofarticles,conference
proceedingsandbookchapters,indicatedatleast57newstudies(i.e.,thosepublishedin
peer-reviewedjournals)thatwerefocusedonsomeaspectofprosodyorprosodic
processinginthespeechofindividualswithASD.Thiskindofattentionissurprisinggiven
thatautismis,afterall,characterizedprimarilybydeficitsinsocialcommunicationand
interactionaswellasbyrestrictedandrepetitivepatternsofbehaviorthataredisabling
(AmericanPsychiatricAssociation,2013).ThediagnosticcriteriasetforthintheDiagnostic
Classification and evaluation of child speech
6
andStatisticalManualofMentalDisorders(DSM-5)donotevenmentionatypicalspeech
soundpatterns,letaloneprosody.Theydohoweverreferencebehaviorssuchas“failureto
initiateorrespondtosocialinteractions”and“abnormalitiesineyecontactandbody
language...toatotallackoffacialexpressions”aswellasstereotypedorrepetitivemotor
movements…orspeech”(e.g.,echolaliaoridiosyncraticphrases).Giventhedisabling
natureofthecriterialbehaviorsinautism,whicharesocialorperseverativeinnature
ratherthanprosodic,isthereagoodreasontofocusevermoreresearchontheunusual
speechpatternsofindividualswithASD?Thisquestionmotivatedthecurrentstudy.
TheCurrentStudy
Thepracticedclinicianapparentlyhasnotroubleidentifyingindividualswith
autismbasedonatypicalspeechsoundpatternswhenthesearepresent(Klin,Sparrow,
Marans,Carter,&Volkmar,2000).Cliniciansintrainingcanalsoreliablydistinguish
speakerswithandwithoutautismbasedonshortconversationalspeechsamplesthatare
measurablydifferentwithrespecttopitchvariability(Nadig&Shaw,2012).Theseabilities
areperhapsnotsurprisinggiventhat,aslanguagespeakers,weareattunedtosound
patternsinspeechthatindexsocialinformation.Thissensitivityallowslistenerstoadeptly
categorizespeechbyethnolect(Lass,Almerino,Jordan,&Walsh,1980;Purnell,Idsardi,&
Baugh,1999),dialect(Clopper&Pisoni,2004;Clopper&Bradlow,2009),genderidentity
(Lassetal.,1980;Munson,2007),andsexualorientation(Munson,McDonald,DeBoe,&
White,2005;Munson,2007).Moreover,listenersdonotthemselvesneedtobeexpertin
thesocialcategoriesofinteresttoeffectivelyclassifyspeechbasedonsoundpattern
differences(Purnelletal.,1999;Clopper&Bradlow,2009).Forexample,Clopperand
Bradlow(2009)usedafreeclassificationtasktoshowthatnon-nativespeakerswereable
toclusterAmericanEnglishspeechsamplesbydialect,albeitlessaccuratelythannative
Classification and evaluation of child speech
7
speakersofAmericanEnglish.Otherstudiesstronglysuggestthattheperceptionofsocial
informationinspeechhassocialconsequences(vanBezooijen,1995;Purnelletal.,1999).
Forexample,Purnellandcolleaguesfoundthatlandlordsdiscriminatedagainsttenants
basedsolelyonspeechpatternsthatindexedanAfricanAmericanorChicanoidentity.
Ourautomaticidentificationofsocialinformationinspeechcoupledwith
impressionisticdescriptionsofatypicalspeechsoundpatternsinautismastooloud,shrill,
sing-songy,mechanicalandsoon(see,e.g.,Asperger,1944inFrith,1991;Klinetal.,2000)
suggestthatautisticspeechmaybenoticeabletoandsociallyevaluatedbylistenerswho
havenoknownpriorexperiencewithautism.Butthequestionremainsastohowimpactful
thesespeechdifferencesreallyare.Morespecifically,isitthespeechpersethatidentifiesa
personwithautismas“other”orisitperhapsthesuiteofunusualcommunicativeand
repetitivebehaviorstypicalofautismthatengendertheperceptionofotherness?Also,if
othernesscanbediscernedbasedsolelyonspeechpatterns,wouldsuchanidentification
resultinaninherentlynegativesocialevaluation?Tobegintoanswerthesequestions,the
currentstudytestedtheeasewithwhichlaylistenerscandistinguishchildrenwithASD
fromtheirtypicallydevelopingpeersbasedsolelyonshortsnippetsoftheirprosodically
impairedspeech.Wealsoinvestigatedhowlistenersevaluatedthisspeechabsentasocial
orlinguisticcontext.
Acousticstudiesthatfindgroupdifferencesbetweenautisticandneurotypical
speechstronglysuggestthatspeech-specificdifferencesarethereandshouldbe
perceptibleeventolaylisteners.Yet,eventhissuggestionhasyettoberigorouslytested.
TheperceptualratingstudyconductedbyNadigandShaw(2012)providessomeevidence
infavoroftheperceptualsaliencyofdifference,buttheirlistenerswerespeech-language
pathologistsintrainingwhowerealsospecificallytaughttoattendtopitchchangesinthe
Classification and evaluation of child speech
8
speechsamplesfromchildrenwithandwithoutASD.Inaddition,thesamplesthatNadig
andShaw’slistenersratedwereexcerptedfromconversationswithnocontrolover
languagedifferences(wordchoice,appropriatenessofresponse,etc.)thatcouldinfluence
judgmentsoftypicality.Thecurrentstudyprovidesamorerigoroustestofthehypothesis
thattheatypicalspeechproducedbychildrenwithASDandunusualprosodyis
perceptuallydistinctfromtypicalchildren’sspeech(Experiment1).Aftershowingthatitis,
wegoontoinvestigatewhetherperceiveddifferenceresultsinanegativesocialevaluation
ofthespeaker(Experiment2).Wefollowupbyidentifyingtheperceptualdimensionsthat
listenersusebothtodistinguishatypicalfromtypicalspeechandtosociallyevaluatea
speaker(Experiment3).Themethodsweusecontrolasbestaspossibleforinfluencesof
social-interactionalcontextonlisteners’classificationandevaluationofspeech:weelicited
children’sspontaneousspeechinanarrativetask,ratherthaninaconversationalone,and
furtherdecontextualizedeventhisspeechbyextractingonlyafew,fluentlyproduced
utterancesfromeachnarrative.Thestimuliandjudgmenttaskswerealsodesignedtotest
thefundamentalsaliencyofprosodicdifferences,andtocontrolforpossibleeffectsof
languageonlisteners’evaluationofspeechpatterns.Finally,weusedlaylisteners,whohad
noknownaprioriexperiencewithASDandrelativelylittleexperienceonaveragewith
children,tomaximizethepossibilitythatlistenerjudgmentswerereflectiveofbroader
societalnormsratherthandependentonpersonalhistoriesandexperiences.Intheseways,
wesoughttoanswerourmotivatingquestionofwhethertheatypicalspeechsound
patternsoftenassociatedwithASDarethemselvesimportanttoaddressdirectlygiventhe
manyothersocially-disablingbehaviorsmorecentraltoautism.
Classification and evaluation of child speech
9
EXPERIMENT1
ThegoalofExperiment1wastotestlaylisteners’abilitytodistinguishchildren
withASDfromtheirtypicallydevelopingpeersbasedsolelyontheiratypicalspeech.The
hypothesisthatlistenerswouldbeabletodothisfollowsfromourautomaticidentification
ofsocialinformationinspeechandfromacousticstudiesthathavemeasuredandfound
differencesinspeechproducedbymanychildrenandadultswithASDcomparedtothose
withoutASD,particularlyintheprosodicdomain.Sincethesestudiesalsoshowthat
prosodicdeficitsdonotappearinallindividualswithASD,weobtainedspeechsamples
onlyfromchildrenwithASDwhowerealreadydeemedbytheirspeech–language
pathologisttohaveunusualspeechprosody.
Methods
Participants
Sixteencollege-agedadultsprovidedlayjudgmentsofdisorder.Theseadultswere
recruitedfromintroductorypsychologyandlinguisticsclasses,andreceivedcoursecredit
fortheirparticipation.AllreportednormalhearingandEnglishastheirnativelanguage.
Thelistenersevaluatedspeechobtainedfrom34children,whoalsohadnormalhearing
andEnglishastheirnativelanguage.Seventeenofthechildren(14boys)hadspecial
education(SPED)eligibilitiesforreceivingspeechandlanguageservicesintheschools
underautismspectrumdisorder(ASD).Theirmeanagewas9;0years(SD=18months).
Theother17children(13boys)weretypicallydeveloping.Theirmeanagewas8;9years
(SD=15months).
ChildrenwithASDwererecruitedforthestudythroughalocalnetworkofschool
speech–languagepathologists(SLPs)in2011.TheSLPswereaskedtoreferchildrenfrom
theircaseloadwhohad(1)aSPEDeligibilityunderASD,(2)normalcognition,and(3)
Classification and evaluation of child speech
10
unusualprosody.InthestateofOregon,adeterminationofautismforSPEDeligibilityis
madebytheschooldistrictandinvolvesateamofeducationprofessionalsincludinga
schoolSLPandclinicalpsychologist.Thisteammakestheirdeterminationbasedon
observationofthechildinandoutoftheclassroom,andusesassessmentinstrumentsto
evaluatetargetbehaviors,cognition,speechandlanguage(seeOregonAdministrative
RulesforSpecialEducationathttp://www.ode.state.or.us/offices/slp/spedoars.pdf).The
AutismDiagnosticObservationSchedule(ADOS;Lord,Rutter,Pamela,Dilavore,&Risi,
2008)isnotused,norisamedicaldiagnosisofautismrequiredoracceptedasasubstitute
forafullevaluationbythedistrictteam.Regardingunusualprosody,wereliedonthe
schoolSLP’sinterpretationofwhatthismightmeaninkeepingwithboththegoalsofthe
studyandtheverybroaddefinitionofwhatconstitutesprosodyinresearchonautistic
speech.
Wecollecteddatafrom18childrenwhowerethuslyreferred,butonechild’sspeech
wasaccidentallynotrecorded.Symptomseverityfortheremaining17childrenwas
measuredbyaskingthereferringSLPtocompletetheChildhoodAutismRatingScale
(CARS2;Schopler,VanBourgodien,Wellman,&Love,2010).Parentalreportwasusedto
establishwhetherornotthechildhadexperiencedlanguagedelay.Receptivevocabulary
scoreswereobtainedusingthePeabodyPictureVocabularyTest(PPVT-4;Dunn&Dunn,
2007).Table1reportsthesex,age,ASDsymptomseverity,languagedelay,and
standardizedreceptivevocabularyscoresforeachchild.
InsertTable1abouthere.
Typicallydevelopingchildrenwerenotspecificallyrecruitedforthecurrentstudy.
Instead,thesechildrenrepresentedasubsetof100childrenwhohadparticipatedina
Classification and evaluation of child speech
11
longitudinalstudyonthetypicalacquisitionofprosodyinschool-agedchildren,whichwas
on-goingatthetimeofthecurrentstudy.Speechsampleswereselectedfrom17ofthe
childrenwhobestmatchedtheASDgroupinageandsexatthetimeofthepresentstudy.
Typicaldevelopmentwasassessedbasedonparentreportregardingthechild’shearing,
speech,language,andmedicalhistory.Themeanageinthisgroupwasnotsignificantly
differentfromthemeanageinthegroupofchildrenwithASD.Table2reportsthesex,age,
andstandardizedreceptivevocabularyscoresofthetypicallydevelopingchildren.
InsertTable2abouthere.
ElicitationProcedure
Structuredspontaneousspeechsampleswereobtainedusingastorytellingtask.An
experimenterpresentedchildrenwith4picturebooksthatdepicteddifferentadventures
ofafrogand/oraboyandadog(i.e.,thefrogstorybooksbyMercerMayer).Childrenwere
askedtochoosethepicturebookthats/hewouldmostliketonarrate.Typicallydeveloping
childrenfamiliarizedthemselveswiththeirbookofchoiceunderexperimenter
supervision,thentoldtheirstorytoaparentorcaregivertwotimes.Storyrepetitionwas
usedtocontrolfortheeffectsoflanguageplanningonthespeechproduced(seeRedford,
2013).Asourinterestisinspeechandnotincognitiveinfluencesonlanguageproduction,
thesamplesusedinthepresentstudywereextractedfromthechildren’ssecond
storytelling.
PilotworkindicatedthatchildrenwithASDresistedtellingthesamestorytwicetoa
parentorcaregiver.Theelicitationprotocolwasthereforeadaptedinthefollowingway.
Childrenchoseabooktonarrate,thendevelopedastoryinresponsetoquestionsand
promptsfromtheexperimenterwhilelookingthroughtheirbookofchoice.Thechild’s
Classification and evaluation of child speech
12
caregiverwastheninvitedbackintotheroom(ortheconversationiftheyhadnotleft),and
thechildwasinstructedto“telltheirstory”tothecaregiver.Thescaffoldingprocedure
servedthesamefunctionasafirststorytellinginourstudyontypicallydeveloping
children’sspeechacquisition:itallowedchildrentoplantheirstoryandpracticethe
languagetheywouldusetotellthestory.
Allchildren’sstoriesweredigitallyrecordedontoaMarantzPMD660(witha
samplingrateof44,100Hz)usingaShureULXS4standardwirelessreceiverandalavaliere
microphone,whichwasattachedtoabaseballhatorheadbandthatthespeakerwore.
ChildrenwithASDwouldsometimesrefusetowearthehatorheadband,inwhichcasethe
microphonewasplacedeitheronthechild’sclothingoronthetablebetweenthe
experimenterandchild.
SpeechSamples
Fourmultiwordutteranceswereextractedfromthemiddleofeachchild’sstoryto
avoidthestereotypedlanguageandprosodyassociatedwithstorybeginningsandendings.
Utteranceswereselectedonthebasisofprosodiccompletenessandlength:utterancesthat
wereroughly2secondslongwereselectedoverthosethatwereshorterormuchlonger.
Thegoalwastoselectphrasesthatwouldbestrepresentphrase-levelrhythmand
intonation(hencetherelativelylongminimumlength)whilecontrollingforsignificant
juncturesduetopausesandotherdisfluencies(hencetherelativelyshortmaximum
length).SamplesfromchildrenwithASDrangedinlengthfrom1.12to3.37seconds(M=
2.21)andfrom4to14syllables(M=8.08).Samplesfromtypicallydevelopingchildren
rangedinlengthfrom1.45to3.06seconds(M=2.17)andfrom5to14syllables(M=9.15).
Onceutteranceswereextracted,filteringwasusedtocreatestimuliforapure
prosodycondition.Theunalteredutteranceswereusedtocreatestimuliforthe
Classification and evaluation of child speech
13
correspondingfullspeechcondition.Thegoalofthemanipulationwastotesttheextentof
prosodicdifferenceinthespeechofASDversustypicallydevelopingchildren,keepingin
mindthatfilteringalwayssubstantiallydegradesthespeechsignal.
Twodifferentbandpassfilterswereused:onepreservedlowerfrequencies(250to
750Hz)andtheotherhigherfrequencies(500to1000Hz).Thebandpassfilterthat
preservedlowerfrequencieswasusedtocapturerhythmicinformationassociatedwiththe
alternationofsonorantandnonsonorantspeechintervals.Thebandpassfilterthat
preservedhigherfrequencieseliminatedtheremaininglowfrequencyenergyassociated
withthosesonorantconsonantsproducedwithsignificantoralocclusioninonsetposition.
Thegoalofthisfilterwasthereforetopreserveinformationaboutthespacingofvowel
onsets,whichhavebeenreferredtoastheperceptualcentersofrhythmicbeatsinspeech
(Port,2003).Notethatpitch(i.e.,F0)informationwaspreservedunderbothmanipulations
becauselistenerscalculatepitchfromthefrequencyspacingbetweenpreservedharmonics
(Moore,1994).Intheend,filtertypehadnosignificanteffectonlistenerjudgmentsandso
werecombinedintheanalysesreportedbelow,whichthereforeonlyrefertofilteredand
unalteredspeechsamples.
Filteredandunalteredutteranceswereamplitudenormalizedto75dBusingthe
scalefunctioninPraat(Boersma&Weenik,2009).Individualutterancesfromeachspeaker
werethenconcatenatedinrandomordertocreatestimulithatwereblockedbyspeaker.
Utteranceswereseparatedby300millisecondsofsilenceineachblock.Blockingwas
deemednecessarybecausepilotworkindicatedthatnaïvelistenersneedtohearmorethan
2secondsofspeechfromasingleindividualtomakeareliablejudgment.Theblocks
rangedindurationfrom7.27to12.51seconds(M=9.80)forchildrenwithASDandfrom
8.18to10.86seconds(M=9.51)fortypicallydevelopingchildren.
Classification and evaluation of child speech
14
ClassificationTask
Listenerswereaskedtomakejudgmentsof“typical”and“disordered”basedonthe
speechsamples.Listenersalsoindicatedtheirconfidenceineachjudgmentona5-point
ratingscale.Allresponseoptionsweredisplayedaslabeledbuttonsonacomputer
monitor.Thetwoalternativejudgments,“typical”and“disordered”,wereplacedinarow
abovetheconfidenceratingscale,whichalsoappearedasarow.Listenerswouldheara
stimulusblock,makeajudgment,ratetheirjudgment,thenpress“OK”tocontinueinthe
task.Stimulusblockswerepresentedinrandomorder.Listenersheardonelow-frequency
bandpassfilteredstimulusblockandonehigh-frequencybandpassfilteredstimulusblock
perspeaker,andsomadeatotalof68typicalityjudgments(34speakersx2repetitions)in
thefilteredspeechcondition.Listenersheardthesameunalteredstimulusblocktwice,and
soalsomadeatotalof68typicalityjudgmentsinthefullspeechcondition.
Listenerinstructionsincludedinformationaboutthespeakers,abouthowthe
speechstimuliwerecreated,andtheeffectofbandpassfilteringonspeech.Specifically,
listenersweretoldthatthespeechsampleshadbeenelicitedfromchildrenbetweenthe
agesof5and11yearsold,andthathalfofthechildrenwerereceivingspeech-language
servicesintheschool.Listenerswereassuredtheresearchobjectivewastohelptarget
speech-languagedeficitsinchildrenwithdisability,andsotheyshouldnotfeelinhibitedin
assigningajudgmentof“disordered”toanyspeechsampletheythoughtwasproducedbya
childwithdisability.Listenerswerealsotoldthattheutterancesthatmadeupeachsample
weretakenfromalargerstorytellingtask,andthatthestorycontextwasnotpreserved.
Finally,listenersweretoldthatsomeofthespeechtheywouldhearwasfilteredtoremove
asmuchlanguageinformationaspossible.Theywereinstructedtomakejudgmentsbased
onwhatremainedperceptibleinthefilteredspeech;namely,rhythmandintonation.The
Classification and evaluation of child speech
15
experimenteralsomadesurethatlistenersunderstoodwhatspeechrhythmandintonation
were.Theseinstructionsweregiventoprovidelistenerswitheveryopportunityto
distinguishchildrenwithASDfromtheirtypicallydevelopingpeerswhenprosodyalone
waspreservedinthefilteredsignal.
Finally,inordertogaininsightintothebasisforjudgmentsof“typical”versus
“disordered,”weaskedlistenerstoprovidewrittenfeedbackregardingtheirperceptual
criteria.Specifically,listenerswereaskedto:“Pleasedescribethecharacteristicsinthe
speechsamplesthatledyoutodistinguishbetween“typical”and“disordered”speech
(please,especially,notewhatyouthinkmadethespeechsounddisordered).”Listeners
respondedtothisrequesttwice:firstaftercompletingjudgmentsinthefilteredspeech
condition,andthenagainaftercompletingjudgmentsintheunalteredspeechcondition.
ControlTasks
Twoadditionaltaskswereincludedintheexperiment;bothwereusedtocontrolfor
theeffectsoflanguage(e.g.,wordchoice,syntacticstructure,samplecoherence)on
“typical”and“disordered”judgments.Thecontroltaskforthefilteredspeechcondition
investigatedtheeffectivenessofthefilteratremovinglanguageinformation:listeners
indicatedroughlyhowmanywordstheythoughttheyunderstoodinthefilteredspeech
samplesona5pointscale:none,some,half,most,orall.These5responseoptionswere
presentedinarowofboxesonthemonitor.Confidenceratingswereagaincollectedon
eachjudgmentthatthelistenermade.Thecontrolfortheunalteredspeechcondition
directlyinvestigatedeffectsofwordchoice,syntacticstructure,andotherlanguage
featuresontypicalityjudgments.Thistaskrequiredlistenerstomake“typical”or
“disordered”judgmentsbasedonwrittenversionsoftheutterancesratherthanonaurally
presentedversions.Thewrittenversionsofaspeaker’sutterancesweredisplayedfor10
Classification and evaluation of child speech
16
secondstoapproximatethedurationoftheaudiosamples.Listenersworeheadphones
whenmakingtext-basedjudgmentsjustasinallotherconditions,theythenclickedon
“typical”or“disordered”asintheotherclassificationtasks,andalsoindicatedtheirdegree
ofconfidenceintheirjudgment.
Theexperimentalandcontroltaskswerealwayscompletedinthefollowingorder:
filteredspeech,comprehension,text(i.e.,writtenstimuli),unalteredspeech.Thefixed
orderhadtodowiththefactthatfilteringonlyeffectivelyeliminatesaccesstolanguage
contentiflistenershavenoaprioriexpectationsaboutthiscontent.
Analyses
Thebinaryresponseswerecoded“0”for“typical”and“1”for“disordered”and
summedacrosslistenerstoobtainasinglescoreforeachsampleineachofthetaskswhere
thisresponsewaselicited.Judgmentscoresintheexperimentaltaskswerethenstrippedof
languageinfluenceinthefollowingway.Standardizedcomprehensionratingscoresfor
eachsamplewereusedtopredictthesummed“disordered”scoresinthefilteredspeech
condition(i.e.,acrossfiltertype)usinglinearregression.Similarly,summed“disordered”
scoresfromthetext-onlyjudgmenttaskwereusedtopredictthesummed“disordered”
scoresintheunalteredspeechcondition.Theresidualscoresfromtheseanalyseswere
saved,andthetwoscoresperspeakeraveragedtoobtainthedependentvariableforthe
principleanalysis,whichinvestigatedlisteners’abilitytodistinguishspeechproducedby
childrenwithASDfromspeechproducedbychildrenwithtypicaldevelopment.Listeners’
confidenceintheirtypicalityjudgmentswerealsoanalyzedasafunctionofspeakergroup.
Thesewerestandardizedwithinlistenerusingaz-transform,thenaveragedacross
listenersandwithinspeakertoobtainthemeanlistenerconfidenceratingperspeaker.
Classification and evaluation of child speech
17
Theprincipleanalysesusedamixed-designANOVAtoassessthebetween-subjects
effectofspeakergroup(ASDversustypical)andwithin-subjectseffectofspeechcondition
(filteredversusunaltered)onthedependentvariables:residualized“disordered”
judgmentsandstandardizedconfidenceratings.AShapiro-Wilktestindicatedthatvalues
associatedwiththesedependentvariableswerenormallydistributed.Asignificanteffectof
groupon“disordered”judgmentswouldindicatethatlistenerscandistinguishchildren
withASDfromtheirtypicallydevelopingpeersbasedonspeechpatternsalone.An
interactionbetweengroupandspeechconditionwouldindicatetheinfluenceofnon-
prosodicfactorson“disordered”judgments.Speaker’sage-in-monthsandstandardized
PPVTscoreswereincludedascovariatesintheanalysestocontrolfortheireffectsonthe
dependentvariables.
Results
ListenerResponsesbyJudgmentTaskandSpeakerGroup
Theuntransformedandunstandardizedresponsesacrossthedifferenttaskswere
notthefocusofanalysis,butaresummarizedinTable3togivethereaderafullersenseof
thedata.Notethatcumulative“disordered”scoreswerehigheronaverageforchildren
withASDcomparedtotheirtypicallydevelopingpeersacrossalltasks.Notealsothatthe
meandifferencebetweenthetwogroupsofchildrenwasespeciallylargeintheunaltered
speechtask.
InsertTable3abouthere.
EffectofSpeakerGrouponResidualizedJudgmentsofDisorder
Recallthattoinvestigatetheinfluenceofspeechpatternsonjudgmentsabsentthe
influenceoflanguageweregressedthescoresobtainedfromthecontroltasks
Classification and evaluation of child speech
18
(comprehensionandtext-basedjudgments)againstthescoresobtainedfromthe
experimentaltasks(filteredspeechandunalteredspeechjudgments).Theanalysisonthe
residualsindicatedthatlistenersweremorelikelytojudgespeechproducedbychildren
withASDas“disordered”thanspeechproducedbychildrenwithtypicaldevelopment,F(1,
30)=4.56,p=.041,ηp2=.132.Thiseffectofgroupinteractedwithcondition,F(1,30)=
4.52,p=.042,ηp2=.131,inthatlistenerswerebetterabletodistinguishbetweenASDand
typicallydevelopingchildrenintheunalteredconditionthaninthefilteredcondition(see
Figure1).Thesimpleeffectofconditionwasnotsignificant.
InsertFigure1abouthere.
Giventhesignificantinteractionbetweenconditionandgroup,andtheseemingly
weakeffectofgroupon“disordered”judgmentsinthefilteredspeechcondition(Figure1),
followupanalyseswereconductedtodeterminewhetherlistenerswereinfactableto
distinguishchildrenwithASDfromtheirtypicallydevelopingpeersinthefilteredspeech
condition.Thedataweresplitbyconditionandtheeffectofgrouponresidualized
“disordered”judgmentswastestedagain.Theresultswerethatlistenersdistinguished
betweengroupsofchildrenbasedonspeechpatternsintheunalteredspeechcondition,
F(1,30)=5.99,p=.020,ηp2=.166,butnotinthefilteredspeechcondition,F(1,30)=0.70,
p>.1.Notethattheeffectofgroupon“disordered”judgmentsintheunalteredspeech
conditionissignificantevenwhenalphaiscorrectedformultiplecomparisons(i.e.,the
Bonferronicorrectedalphais.025).
WrittenFeedbackfromListenersontheirJudgments
Again,theeffectsoflanguagewerecontrolledintheunalteredspeechconditionby
partiallingoutvarianceinjudgmentsduetowordchoice,syntacticstructure,andso.Thus,
Classification and evaluation of child speech
19
theresultsshowninFigure1indicatethatlistenersdifferentiatedchildrenwithASDand
prosodicdisorderfromtheirtypicallydevelopingpeersbasedonspeechsoundpatterns
alone.Yet,therewasnoeffectofgroupontheresidualized“disordered”judgmentsinthe
filteredspeechcondition.Thissuggestseitherthatprosodicdifferencesbetweenthetwo
groupsofchildrenweresubtleatbestandlistenersonlygainedsensitivitytothose
differencesastheexperimentprogressedorthatlistenersattendedtosomethingother
thanprosodywhenmakingjudgmentsofdisorder.Listenerfeedbacksuggestedboth
explanationscouldbetrue.Specifically,listenersreportedattendingtoarticulation,speech
clarity,fluency(=flow,rate,rhythm),andaccenting(=wordemphasis,intonation)when
makingjudgments.Fluencyandaccentingareclearlyprosodicinnature;articulationand
speechclarityreferencesegmentalcharacteristics.Table4summarizesthefeedbackwe
obtainedfromeachofthe16listenerswhoprovidedjudgmentsonthefilteredand
unalteredspeechsamples.
InsertTable4abouthere.
Listeners’ConfidenceintheirJudgments
Theanalysisonlisteners’standardizedconfidenceratingssupportthesuggestion
thatthespeechsamplesobtainedfromchildrenwithASDweresubtlydifferentfromthose
obtainedfromtypicallydevelopingchildren.Theymayalsohavebeenmorevariable.
Specifically,therewasasignificanteffectofgrouponlisteners’standardizedratings,F(1,
30)=10.75,p=.003,ηp2=.264,asshowninFigure2.Listenerswerelessconfidentintheir
judgmentsofspeechproducedbychildrenwithASDthanintheirjudgmentsofspeech
producedbychildrenwhoweretypicallydeveloping.Therewasnoeffectofconditionor
anyinteractionbetweenconditionandgroup.
Classification and evaluation of child speech
20
InsertFigure2abouthere.
Discussion
TheresultsfromExperiment1indicatethatlistenerscorrectlyidentifyspeech
producedbychildrenwithASDas“disordered”moreoftenthanspeechproducedby
typicallydevelopingchildren.Thesoundpatterndifferencesaresufficientlysalientthat
listenerscandistinguishthegroupsoutsidethesocial-interactionalcontextandbasedon
veryshortspeechsamples.Yet,theabsenceofagroupeffectonjudgmentsinthefiltered
speechconditionsuggeststhatthesoundpatterndifferencesaresubtleenoughtobe
obscuredindegradedspeech.Thissuggestionisfurthersupportedbylisteners’weaker
confidenceintheirratingsofatypicalchildren’sspeechcomparedtotypicalchildren’s
speech.Listeners’weakerconfidenceintheirratingsofatypicalspeechmayalsoindicate
thatthesamplesprovidedbychildrenwithASDweremorevariablethantheonesobtained
fromtheirtypicallydevelopingpeers.
EXPERIMENT2
ThegoalofExperiment2wastotestwhetherlisteners’socialevaluationofspeakers
basedonspeechsoundpatternsalsodifferentiateschildrenwithASDfromtheirtypically
developingpeers.Thisgoalisinserviceofourquestionaboutwhetheratypicalspeech
soundpatternsmatter,wherethenotionofwhatmattersisunderstoodasaquestionabout
theeffectsofdifferenceonlisteners’attitudetowardsthespeaker.Therelationship
betweenlaylisteners’judgmentsofdisorder(Experiment1)andtheirlikeabilityratings
wasalsodirectlyexplored.
Classification and evaluation of child speech
21
Methods
Participants
Twelvenewcollege-agedadultsprovidedlikeabilityratingsonthespeechsamples
fromExperiment1.Theseadultswererecruitedfromintroductorypsychologyand
linguisticsclasses,andreceivedcoursecreditfortheirparticipation.Allreportednormal
hearingandEnglishastheirnativelanguage.
RatingTask
ListenersratedboththefilteredandtheunalteredspeechsamplesfromExperiment
1forlikeabilityona7-pointLikertscale.Thelowvalueonthescalewasanchoredwitha
positivesocialstatement(“1=Awesome!Lovethiskid.”)andthehighvaluewithanegative
socialstatement(“7=Nah.Wouldn’tlikethiskid.”).Listenerssatinfrontofacomputer
monitorwiththeanchorsdisplayedabovetheratingscale,whichwasrepresentedasa
numberedsequenceof7buttons.Theylistenedtoaspeechsampleoverheadphones,rated
italongthelikeabilityscale,clicked“OK”andadvancedtoanewscreenandanewsample.
AsinExperiment1,listenerscompletedthefilteredspeechconditionbeforetheunaltered
speechcondition.Speechsampleswereplayedindifferentrandomorders.Listenerswere
toldonlythatthechildrenwerebetween5and11yearsold.UnlikeinExperiment1,they
werenottoldthathalfofthechildrenwerereceivingspeechandlanguagetherapyinthe
schools.
Analyses
Likeabilityratingswerestandardizedwithinlisteneracrossconditionsby
convertingtheratingscoresintoz-scores.Ratingswerethenaveragedacrosslistenersand
withinspeakers.Amixed-designANOVAwasusedtoassessthebetween-subjectseffectof
speakergroup(ASDversustypicaldevelopment)andwithin-subjectseffectofspeech
Classification and evaluation of child speech
22
condition(filteredversusunaltered)onlikeabilityratings.Asignificanteffectofgroupon
likeabilityratingswouldindicatethatlistenersevaluatechildrenwithASDdifferentlyfrom
theirtypicallydevelopingpeersbasedonspeechpatternsalone.Aninteractionbetween
groupandspeechconditionwouldindicatetheinfluenceofnon-prosodicfactorson
listeners’ratings.Speaker’sage-in-monthsandstandardizedPPVTscoreswereincludedas
covariatesintheanalysestocontrolfortheireffectsonthedependentvariable.
Asecondanalysisusedlinearregressiontoinvestigatetherelationshipbetweenthe
residualized“disordered”judgmentsfromExperiment1andthelikeabilityratings
obtainedfromadifferentgroupoflistenersinExperiment2.Likeabilityratingswerethe
dependentvariable,andjudgmentsofdisorderthepredictorvariable.Age-in-months,
PPVTscores,andspeechconditionwereenteredascontrolpredictorvariables.
Results
EffectofSpeakeronLikeability
Theanalysisindicatedasignificantinteractionbetweengroupandspeechcondition,
F(1,30)=5.17,p=.030,ηp2=.147,whichisshowninFigure3.Neitherthesimpleeffectof
groupnorthatofspeechconditionwassignificant,thoughtheeffectofgroupapproached
significance(p=.051).
InsertFigure3abouthere.
Giventheinteraction,theeffectofgroupwasexploredfurtherinanalysessplitby
condition.Theseanalysesindicatedasignificanteffectofgrouponlikeabilityratingsinthe
unalteredspeechcondition,F(1,30)=6.79,p=.014,ηp2=.184,butnotinthefiltered
speechcondition.AsshowninFigure3,childrenwithASDwereperceivedassignificantly
lesslikeablethanchildrenwhoweretypicallydevelopingbasedontheunalteredspeech
Classification and evaluation of child speech
23
samples1.Notethattheeffectissignificantevenwhenalphaiscorrectedformultiple
comparisons(i.e.,theBonferronicorrectedalphais.025).
RelationshipbetweenJudgmentsofDisorderandLikeability
TheresultsonlikeabilityrecallthosefromExperiment1inthatlistenersonly
discriminatedbetweenthegroupsofspeakersintheunalteredspeechcondition.The
similaritybetweenthetwoexperimentswasconfirmedinalinearregressionanalysis,
whichshowedthattheresidualizedjudgmentsofdisordercouldbeusedtoexplain/predict
likeabilityratings,β=.59,t(67)=6.01,p<.001,evenwithage-in-months,PPVT,and
speechconditioncontrolledinthemodel.Byitself,thepredictor“disordered”judgments
accountedfor42%ofthevarianceinlikeabilityratings.Noneofthecontrolvariables
explainedasignificantproportionofthevariance.
TherelationshipbetweenperceiveddisorderandlikeabilityisshowninFigure4.
Speechsamplesthatelicitedahigher“disordered”responseelicitedamorenegativesocial
evaluation(“Nah.Wouldn’tlikethiskid.”).Whensplitbyspeechcondition,wefindthatthe
correlationbetweenthejudgmentsofdisorderandlikeabilityratingswasstrongerinthe
unalteredspeechcondition,r(34)=.72,p<.001,comparedtothefilteredspeechcondition
r(34)=.48,p=.004,thoughthisdifferencewasnotsignificant,z=1.51,p=.066(one-
tailed).
1Recallthatthe7-pointlikeabilityscalewasanchoredsuchthathigherscoresindicateamorenegativesocialevaluationthanlowerscores(“1=Awesome!Lovethiskid.”versus“7=Nah.Wouldn’tlikethiskid.”).
Classification and evaluation of child speech
24
InsertFigure4abouthere.
Discussion
TheresultsfromExperiment2indicatethatlaylistenersevaluatechildrenwithASD
morenegativelythantheirtypicallydevelopingpeers.Thus,likeinExperiment1,listeners
wereabletodistinguishbetweenspeakersbasedontheshortspeechsamplesprovided.
TheresultsalsoparallelthosefromExperiment1inthatlistenerswerenotableto
distinguishbetweenthedifferentgroupsofspeakersinthefilteredspeechcondition.A
regressionanalysisconfirmedthatthesimilaritiesbetweenExperiment1andExperiment
2wereduetoasignificantcorrelationbetweenlaylistenersjudgmentsofdisorderand
theirsocialevaluationofchildren.Next,weinvestigatewhetherthiscorrelationmeansthat
thesamesoundpatternstriggerboththeperceptionofdisorderandanegativesocial
evaluation.
EXPERIMENT3
InExperiment3,furtherperceptualevaluationofchildren’sspeechandacoustic
measuresofsegmentalandsuprasegmentalfeatureswereusedtoidentifythespeech
soundpatterningthatmediatedlisteners’judgmentsof“typical”versus“disordered”in
Experiment1andtheirlikeabilityratingsinExperiment2.Thegoalwastoanswerthe
questionofwhetherbothtypesofjudgmentsreferencethesamepatternsinorderto
understandwhyajudgmentofdisordermayleadtoanegativesocialevaluation.Thus,the
focusofExperiment3isonlisteners’behavior.Wedonotinvestigatethisbehaviorasa
functionofthespeakergroups,butdoconfirmthatchildrenwithASDdifferfromtypically
developingchildrenalongtheperceptualandacousticdimension(s)thatpredictjudgments
ofdisorderandtheirlikeabilityratings.
Classification and evaluation of child speech
25
Methods
Participants
Sixteennewcollege-agedadultsratedtheunalteredspeechsamplesalongseveral
dimensions.Theseadultswererecruitedfromintroductorypsychologyandlinguistics
classes,andreceivedcoursecreditfortheirparticipation.Allreportednormalhearingand
Englishastheirnativelanguage.
RatingTask
ListenersratedtheunalteredspeechsamplesfromExperiment1and2ona7-point
Likertscalealongthe5dimensionsderivedfromthesubjectivefeedbackacquiredin
Experiment1:articulation,speechclarity,fluency,andaccenting.Afifthdimension,
monotony,wasincludedtodifferentiateratingsthatmayhavebeenduetodifferences
appropriatelypitchaccentinglinguisticcontent(i.e.,tune-to-textalignment)fromthose
duesimplytopitchvariability.Listenerswereonlytoldthattheywouldberatingchildren’s
speechalongdifferentperceptualdimensions.Listenerswerealsotoldtheagerangeof
childspeakers,butnomentionwasmadeofthefactthatsomeofthechildrenwere
receivingspeechandlanguagetherapyintheschools.Thespeechsampleswereexplained
asbeforewithreferencetotheelicitationmethod.Onlyunalteredspeechsampleswere
usedinExperiment3becausetheseelicitedthehighestcorrelationsbetweenjudgmentsof
disorderandlikeability(seeExperiment2),andbecauseweneededtolimitthenumberof
stimulipresentedtolistenersinorderforthemtoratealldimensionswithinthe45to55
minutesallottedto“workwithbreaks”duringtheexperiment.
Listenerssatinfrontofacomputermonitorwiththeanchorsdisplayedabovethe
ratingscale,whichwasrepresentedasanumberedsequenceof7buttons.Thelowvalue
onthescalewasanchoredwithadescriptionoftypicalspeechandthehighvaluewitha
Classification and evaluation of child speech
26
descriptionofatypicalspeech.Thelowandhighanchorsforarticulationwere“1=ALL
soundscorrectlyproduced”and“7=NOsoundscorrectlyproduced;”forclearness,“1=
ExtremelyClear”and“7=TotallyUnclear;”forfluency,“1=SuperFluent”and“7=Really
Disfluent”.Theanchorsforaccenting2were,respectively,“1=EmphasizesMostImportant
Words”and“7=EmphasizesOnlyIrrelevantWords;”andformonotony,“1=VeryLively”
and“7=TotallyMonotonous.”Listenersheardaspeechsampleoverheadphones,ratedthe
sampleonthescalealongthespecifieddimension,clicked“OK”andadvancedtoanew
screenandanewsample.Thereweretworepetitionsofeachsampleasbefore.Allsamples
werepresentedinrandomorder.Theorderinwhichdimensionswereratedwasvaried
acrosslisteners.Fourteenlistenerscompletedratingsalongthearticulationdimension3.All
listenerscompletedratingsalongallotherdimensions.
AcousticMeasurements
Inordertobettercharacterizespeechsoundpatterningrelevanttothedimensions
ofarticulation,speechclarity,fluency,accenting,andmonotony,anumberofacoustic
segmentalandsuprasegmentalmeasuresweretakenonthe136utterancesthatwerethe
speechsamplesprovidedtolisteners.Measurementswerebasedonthehand-
segmentationoftheutterancesintoconsonantandvowelintervalsinPraatusingstandard
segmentationcriteria(see,e.g.,Redford,2014).Voweldurations,F0,F1,andF2were
automaticallyextracted.ThemeasurementintervalforF0wassetat.01secondsandthe
2Theanchorsforratingalongtheaccentingdimensionweremeanttodrawlisteners’attentiontotheintonationalaspectsofthephrase,andespeciallytotheappropriate/inappropriateuseofprosodicfocus,followingtheliteratureonprosodicdeficitsinspeechproducedbyindividualswithASD.3Listenerswererun2atatimeinthelaboratory.Ononeoccasion,theexperimentalsessionstartedlateandsothelistenerswhoparticipatedinthatsessionwerenotabletocompletethetaskintime.Sincetaskorderwasvariedacrosssessions,bothlistenersintheshortenedsessionwereunabletocompletethesame/lasttaskinthesession.
Classification and evaluation of child speech
27
rangesetfrom75to600Hz.ValuesthatdeviatedoneSDfromaspeaker’smean,werere-
measuredbyhand.VowelformantmeasureswerebasedonthePraatLPCalgorithmin
timestepsequalto10percentofthevowelinterval,andwiththemaximumnumberof
formantssetto5andmaximumformantfrequencyto6000Hz.Everyformanttrackwas
visuallyinspected,thenhand-correctedandre-measuredifthetrackswereoff.
Segmentalmeasures.Voweldurationandformantmeasureswereusedtoderive3
measuresofarticulation:schwaduration,meannormalizedvariabilityinschwaduration,
andF1xF2vowelspace.Voweldurationandstabilityinrepeatedvowelproductionare
indicatorsofspeechmotorcontrol,withshorterdurationsandmorestableproductionsa
featureofgreatercontrolovertargetedarticulations(seeRedford&Oh,2017).Schwa
vowelsareminimallyinfluencedbyprosodicfactorssincetheydonotreceivelexicalstress
orphrasalaccents,someanschwadurationandvariabilityindurationislesssensitiveto
contexts.Therewerealsomultipleschwaproductionsineveryspeechsample,soschwa
durationandvariabilitywassomethingwecouldcalculateforallspeakers.TheF1xF2
vowelspacewasdefinedbyfullvowelsthatwerealsomonopthongs.Thespacewas
calculatedasthemeanEuclidiandistanceofeachfullvowelfromthemeanfullvowelin
eachutterance.Thismeasurethusprovidesinformationaboutvoweldistinctivenesswithin
anutterance,whichisacorrelateofspeechclarity(see,e.g.,Lindblom,1990).
Suprasegmentalmeasures.ThedurationandF0measureswereusedtoderive
measuresofrate,rhythmandintonation.Articulationratewascalculatedasnumberof
syllablespersecondofspeech,excludinganysilencesduetopausing,whichwasextremely
rareinanycase(seespeechsamplesunderExperiment1).Thecoefficientofvariationin
voweldurations(i.e.,standarddeviation/mean)wasusedtocapturespeechrhythm.This
measurehasbeenproposedintheliteratureasaneffective,simple-to-compute,rate-
Classification and evaluation of child speech
28
normalizedrhythmmetric(White&Mattys,2007).Finallengthening,afeatureofprosodic
boundarymarking,wascalculatedastheratioofphrase-finalvoweldurationtothe
averagenon-finalvowelduration.Anotherfeatureofprosodicboundarymarking,final
pitchchange,wascalculatedasF0changefromthepenultimatetoultimatesyllableofthe
phrase.Pitchdeclination,acharacteristicofthewholeintonationcontour,wascalculated
astheslopeofabestfitregressionlineplottedthroughsequentiallyarrangedF0valuesfor
eachutterance.
Inadditiontothesegmentalandsuprasegmentalmeasures,severalacoustic
measurescorrespondingtovoiceweretakentodeterminewhetheroverallvoicequality
mighthavecontributedtolistenersjudgmentsofdisorder(see,e.g.,Shribergetal.,2001).
ThemeasuresweremeanF0andtwomeasuresoftheproportionoftimeduringaglottal
periodwherethereisnocontactbetweenthevocalfolds,whichisrelatedtovocalquality
rangingfromvocalfrytobreathy.These“openness”measureswerethemeanamplitude
differencebetweenthefirstharmonic(H1)andthesecondharmonic(H2)acrossallvowel
intervals,andorthemeanamplitudedifferencebetweenH1andthefirstformant(A1)(see
Garellek&Keating,2011).AswiththeF0andformantvalues,phonationmeasureswere
calculatedautomaticallyusingaPraatscript.F0andformantcalculationsweresetas
before.
DataReduction
Ratingsalongeachdimensionwerestandardizedwithinlistenerbyconvertingthe
Likertscoresintoz-scores.Ratingswerethenaveragedacrosslistenersandwithin
speakers.Inter-correlationsbetweentherateddimensionswereassessed.Ratingsalong
thedimensionsofarticulation,clearness,andfluencywereexceptionallyhighlycorrelated,
withcorrelationcoefficientsrangingfrom.92(articulationandfluency)to.96(articulation
Classification and evaluation of child speech
29
andclearness).Accordingly,these3dimensionswerereducedtooneusingprincipal
componentsanalysis(PCA).ScoresfromthefirstfactorofthePCAweresavedandusedin
placeofthearticulation,clearness,andfluencyratings.Foreaseofreference,wewillrefer
tothesescoresascapturingthedimensionofintelligibility.Thisdescriptivelabelwas
chosenbecausetheconstituentcomponentsofthisshareddimension(e.g.,articulation,
clearness,andfluency)areallattributesofintelligiblespeech.Intelligibility,accenting,and
monotonywerethenenteredintoamultipleregressionmodeltopredictjudgmentsof
disorder(seebelow).Thismodeldidnotsufferfromunacceptablecollinearity(VIF<3.5).
Inter-correlationsbetweentheacousticmeasureswerealsoassessed.Manydidnot
reachsignificance.AllandonlytheonesthatdidareshowninTable5.Notethenear
perfectcorrelationbetweenthetwomeasuresrelatedtovoicequalityinthetable.Given
thehighdegreeofoverlap,thesemeasureswerereducedtooneusingPCA.Scoresfromthe
firstfactorofthePCAweresavedandusedinplaceoftheindividualH1-H2andH1-A1
measures.Foreaseofreference,wewillrefertothesescoresasmeasuresofvoicequality.
InsertTable5abouthere.
StatisticalAnalyses
Stepwiselinearregressionmodelingwasusedtoinvestigatetherelative
contributionoflistener-identifiedperceptualdimensionsofspeechsoundpatterningto
residualized“disordered”judgmentsandz-scoredlikeabilityratingsonunalteredspeech
samples.Age-in-monthsandreceptivevocabularyscoreswereincludedascontrol
variablesinthemodel.Backwardeliminationwasusedtoreducethenumberofpredictor
variablestoonlythosethatcontributedsignificantlytoexplaininglisteners’responses.The
Classification and evaluation of child speech
30
sametechniquewasthenusedtocharacterizeinacoustictermstheperceptual
dimension(s)thatbestexplainedlisteners’behaviorinExperiments1and2.
Results
PerceptualBasisforJudgmentsofDisorder
Thefullmodelaccountedfor69%ofthevarianceinjudgmentsofdisorderandhad
anadjustedR2of.63.Thesignificantmodelwiththefewestpredictorsaccountedfor67%
ofthevarianceandhadanadjustedR2of.65.Theonlysignificantpredictorsinthismodel
wereage-in-months,β=.22,t(33)=2.10,p=.045,andintelligibility,β=.84,t(33)=7.98,p
<.001;thatis,theperceptualdimensionderivedfromtheoverlapinratingsofarticulation,
clearness,andfluency.Whereasintelligibilitybyitselfaccountedfor63%ofthevariancein
“disordered”judgments,themodelwithage-in-monthsalonewasnobetterthanthenull
modelataccountingforvariance.Figure5showstherelationshipbetweenpoor
intelligibility4and“disordered”judgments.
InsertFigure5abouthere.
Anindependentsamplest-testconfirmedthatvariancealongtheintelligibility
dimensionalsovariedsystematicallywiththegroupofspeakerswhoproducedthe
samples,MeanDifference=1.02,t(33)=3.40,p=.002(two-tailed).Althoughchildrenwith
ASDwereratedaslessintelligibleonaveragethantheirTDpeers,itisalsoevidentfrom
thedatapresentedinFigure5thatnotallchildrenwereratedasequallyunintelligiblenor
weretheyallratedasequally“disordered”.
4Recallthatthe7-pointratingscalewasanchoredsuchthathigherscoresaremorecompatiblewithdescriptionsofdisorderedspeech.(e.g.,“1=ALLsoundscorrectlyproduced”versus“7=NOsoundscorrectlyproduced”).
Classification and evaluation of child speech
31
PerceptualBasisforLikeabilityRatings
Thefullmodelaccountedfor57%ofthevarianceinlikeabilityratingsandhadan
adjustedR2of.49.Thesignificantmodelwiththefewestpredictorsaccountedfor49%of
thevarianceandhadanadjustedR2of.46.Theonlysignificantpredictorsinthismodel
wereintelligibility,β=.56,t(33)=4.12,p<.001,andmonotony,β=.29,t(33)=2.15,p=
.040.Intelligibilityaccountedfor42%ofthevarianceinlikeabilityratingsbyitself,and
monotonyfor21%.Thus,itwouldseemthatintelligibilityandmonotonyeachaccounted
forsomeofthesamevarianceintheregressionmodel.Figure6showstherelationship
betweenintelligibilityandlikeability,wherehighervaluesalongeachaxisindicatemore
negativeevaluations.
InsertFigure6abouthere.
AcousticCharacterizationofIntelligibilityandMonotony
Segmental,suprasegmental,andvoicemeasureswereenteredaspredictorvariables
inastepwiseregressionmodelstoprovideacousticcharacterizationsoftheperceptual
dimensions,intelligibilityandmonotony.Thefullmodelsaccountedfor46%ofthe
varianceinintelligibilityandfor68%ofthevarianceinmonotony;theyhadadjustedR2sof
.20and.55,respectively.Thesignificantmodelswiththefewestpredictorsaccountedfor
41%ofthevarianceinintelligibilityandfor56%ofthevarianceinmonotony;theyhadan
adjustedR2sof.17and.50,respectively.Articulationratewasthesolesignificantpredictor
ofintelligibility,β=–.41,t(33)=–2.55,p=.016:slowerrateswereassociatedwithpoorer
intelligibility.Bycontrast,thereducedmodelofmonotonyincludedtwosignificant
segmentalmeasures—meanschwaduration,β=.31,t(33)=2.54,p=.016,andvowel
distinctiveness,β=–.25,t(33)=–2.04,p=.050—andtwosignificantvoicemeasures—
Classification and evaluation of child speech
32
meanF0,β=–.37,t(33)=–2.45,p=.020,andvoicequality,β=.44,t(33)=2.91,p=.007.
Thedirectionoftheeffectofschwadurationisconsistentwiththeeffectofspeechrate:
longerschwawereassociatedwithgreaterperceivedspeechmonotonythanshorter
schwas.Theeffectofvoweldistinctivenessindicatesthatmoremonotonousspeechwas
alsolikelylessclear.Lowerpitchandlessmodal(morebreathy)voicequalitieswerealso
associatedwithmoremonotonousspeech.Giventhislistofsignificantpredictorvariables,
itwouldseemthatalthoughtheperceptualdimensionsofintelligibilityandmonotonyare
acousticallyseparable,botharewell-characterizedbyfeaturesthataremoreassociated
withspeechmotorandvoicefactorsthanwithlinguisticprosody.
Discussion
TheresultsfromExperiment3stronglysuggestthatlaylistenersattendtopatterns
relatedtomotorspeechratherthantoprosodiconeswhentheyareaskedtoclassify
children’sspeechsoundpatternsas“disordered”or“typical”andtoevaluatehowmuch
theylikeachildbasedonshortspeechsamples.Thisconclusionisbasedonthefindingthat
“accenting,”themostovertlyprosodicperceptualdimensionevaluated,wasnota
significantpredictorofjudgmentsofdisorderoroflikeability.Moreover,thestrongest
predictorofbothperceiveddisorderandlikeabilitywasintelligibility,adimensionderived
fromtheoverlapinratingsofarticulation,clarity,andfluency.Thisdimensionwasbest
characterizedbyarticulationrate.Ratemaybeasuprasegmentalfeatureofspeech,butitis
onemoretypicallyassociatedwitharticulatorytimingcontrolthanwithprosody(see,e.g.,
Redford,2014).Similarly,thedimension“monotony”wasbestcharacterizedbymeasures
thatcapturedtheaccuracyandspeedofsegmentalarticulations(schwadurationand
voweldistinctiveness)andbymeasuresrelatedtovoicecharacteristics(pitchandvoice
quality)thatarealsooutsidethedomainofprosody,strictlydefined.
Classification and evaluation of child speech
33
GENERALDISCUSSION
AlthoughchildrenandadultswithASDareoftenreportedtohaveatypicalspeech
patterns,manyindividualswithautismshownoprosodicdeficits(seeShribergetal.,2001;
2011;Nadig&Shaw,2012).Moreover,thebehaviorsthatarecentraltoestablishinga
diagnosisofautismaremoreobviouslysociallydisabling(e.g.,notrespondingtoone’sown
name)anddeviant(e.g.,echolalicspeech)thanwemightimagineunusualspeechpatterns
tobe.Thecurrentstudywasmotivatedbythequestionofwhetheratypicalspeechpatterns
inchildrenwithASDisimportantenoughtowarrantresearch.Inaclinicaloreducational
context,thenotionofwhatbehaviorisorisnotimportanttounderstandamountstoa
questionabouttheeffectsthatsuchbehaviorhasonpeerrelationsoronlearning.The
resultsfromthepresentstudyprovideevidenceinsupportoftheviewthatatypicalspeech
soundpatternsmarkchildrenwithASDasdifferentfromtheirTDpeers,andarelikelyto
createbarrierstopositivesocialinteractions.Weconcludefromthisthatatypicalspeech
shouldbeaddresseddirectlyinchildrenwithASD.Theresultsdonot,however,support
theviewthatprosodyshouldnecessarilybethefocusoftheseefforts.Wheninputis
limited,listeners’judgmentsareinfluencedbyspeechpatternsthatlikelyreflectdelayedor
deviantspeechmotorcontrolandunusualvoicequalityratherthanprosodicdeficitsper
se.Morespecifically,thecurrentstudyidentifiesintelligibilityastheprimaryperceptual
dimensionthatmediatesbothlisteners’judgmentsofdisorderandtheirsocialevaluation
ofthechildbasedonshortsnippetsofspeech.Thisdimensioniscloselyassociatedwiththe
perceivedgoodnessandfluency(=speed)ofsegmentalarticulation.
Intelligibilityisaconceptfundamentaltostudiesofmotorspeechdisorders(e.g.,
dysarthria),whereitisusuallymeasuredasthenumberofdistinctspeechsoundsorwords
recognizedbyapaneloflisteners(seee.g.,Kentetal.,1989).Wewouldthereforeliketo
Classification and evaluation of child speech
34
clarifythatnoneofthechildrenwhoparticipatedinthepresentstudysoundeddysarthric.
Theirspeechwasnotdistorted,justdifferentsounding.Thisdifferencewascharacterized
bythereferringSLPsasunusualprosody,butlisteners’behaviorsuggeststhatperhaps
speechmotorfactors,ratherthanlinguisticprosodicones,weremoreimportanttothe
perceptionofdifference.Itisforthisreasonthatwewouldarguethatthenotionof
intelligibilitybettercaptureswhatlistenersweremostsensitivetowhendistinguishing
childrenwithASDfromtheirtypicallydevelopingpeersbasedontheshortspeechsamples
provided.Additionalevidenceforthisviewcomesfromthefeedbackthatlisteners
providedinExperiment1.Forexample,Listeners1,2,and15allnotethattheirjudgments
wereaffectedbywhattheyperceivedas“slurred”or“slurring.”Listeners11and13refer
totheinfluenceofa“lisp”ontheirjudgments.Listeners3,4,5,7,8,11,and15commenton
aspectsofrate.Relatedly,articulationratewasasignificantpredictorofintelligibility.
Elsewhere,wehavefoundelsewherethatdefaultarticulationratesindextheacquisitionof
articulatorytimingcontrolmorebroadly,atleastintypicallydevelopingchildren(Redford,
2014).
Theconclusionthatlistenersmayhavefocusedonspeechmotorfactorsto
distinguishbetweenchildrenwithASDandTDisatoddswiththecharacterizationof
autisticspeechasprosodicallydisorderedifoneunderstandsspeechprosodyinlinguistic
terms;thatis,asasystemofmetricalgroupingandaccentplacementsgovernedbya
prosodicgrammarthatislinked,atsomelevel,tosyntacticstructureandsemantic-
pragmaticmeaning.Deficitsinlexicalstressplacementandcontrastivefocusmeetthis
definitionofprosodyandsoareconsistentwiththeprimarydeficitsinsemantic-pragmatic
communicationthatisassociatedwithASD,butthesewerenotfoundhere.Moreover,the
differenceswedidfindcorrespondtothemanyotherdocumenteddifferencesbetween
Classification and evaluation of child speech
35
autisticspeechandtypicalspeechthatarenotnecessarilyprosodic.Inparticular,acoustic-
phoneticstudiesthatcharacterizedifferencesbetweenASDandtypicalspeechintermsof
globaldifferencesinpitchrangeandvariability(Diehletal.,2009;Nadig&Shaw,2012),
longerwordandutterancedurations(Diehl&Paul,2012),andhigherratesofdisfluencies
(Shribergetal.,2001)areallcompatiblewithdifferencesinplanningandproductionand
notwithprosodyperse.
Thefindingsthatlistenersaresensitivetospeechintelligibilitywhenjudgingshort
snippetsofspeechandthatthissensitivityaffectstheirsocialevaluationofaspeakeralso
hasimplicationsforconceptualizingthesemantic-pragmaticimpairmentsinchildrenwith
ASD.Inparticular,afocusonintelligibilityemphasizestheimportanceofthesocial
interactionfordefining“impaired”,“atypical”or“disordered”,andsosupportsaviewof
pragmaticimpairmentasemergentfromtheinteractionbetweenspeakersandlisteners
(seePerkins,2007).Whenthecontributionsofbothareinbalance,communicationis
successfulandpragmaticsareintact.Whenonememberofthedyadmustadaptinan
unfamiliar,asymmetricalmannertoaccommodatetheother’sbehavior,then
communicationsuffersandpragmaticsisimpaired.Thisviewofimpairmenthasthe
advantageofallowingustounderstandwhylisteners’perceptionofdisordermightaffect
theirsocialevaluationofthespeaker.Anydecrementinspeechintelligibilityislikelyto
resultincommunicativeasymmetry;namely,anasymmetrythatcreatesextraworkforthe
listener.Iflisteners(implicitly)resentdoingextrawork,thentheymayblamethespeaker.
NotethatthisexplanationissimilartothatwhichRiceandcolleaguesproposedtoexplain
theirfindingsthatchildrenwithspecificlanguageimpairmentorwhoaresecondlanguage
speakersofEnglishhavelessaccesstopositivesocialinteractionsthantheirtypically
Classification and evaluation of child speech
36
developingandnativeEnglish-speakingpeersinUSschools(Riceetal.,1991;Gertneretal.
1994).
Althoughwehavesuggestedthatlistenersaremoresensitivetospeechsound
patternsthatsuggestdeficitsinspeechmotorcontrolratherthaninprosody,itis
importanttonotethatthestrengthofthisconclusionislimitedbytheexperimentaldesign
weused.Specifically,theshort,decontextualizedspeechsamplespresentedtolisteners
mayhaveforcedthemtolatchontosomethingotherthanprosodytomaketheirdecisions.
Ifunusualprosodyinautismoftenreflectsinappropriateorincorrectprosodicfocus
(Shribergetal.,2001;Paul,Augustyn,etal,2005;Peppéetal.,2007),thenaccessto
languagecontextwouldbecriticalforlistenerstoidentifytheatypicalpatterns.Infact,one
waytointerpretthefindingthatlistenerscouldonlydistinguishchildrenwithASDfrom
theirtypicallydevelopingpeersintheunalteredspeechconditionisthatonlythis
conditionprovidedlistenerswiththerelevantinformationabouttune-to-textalignment.
Moreover,somelistenersdidcommentthathowaspeakerusedaccentinginfluencedtheir
judgmentsofdisorder.Forexample,Listener4inExperiment1notedthattheirjudgment
wasaffectedbywhetherthechild“stressedtherightsyllables”andListener16was
sensitivetowhetherornotemphasiswasusedcorrectly.Futureworkshouldevaluatethe
relativeinfluencesofintelligibilityandungrammaticalprosodicfocusassignmentand
phrasingonlisteners’behaviorbyusinglongerstretchesofspeechthatprovideadiscourse
context.
Inconclusion,theresultsfromthepresentstudyshowthatlaylistenersare
sensitivetotheatypicalspeechpatternsproducedbychildrenwithASD.Thissensitivity
alsoinfluencestheirsocialevaluationofthespeaker.Speakerswhoproducedpatternsthat
wereperceivedasatypicalordisorderedandratedaslesslikeable.Thesefindingsvalidate
Classification and evaluation of child speech
37
effortstounderstandandtreatatypicalspeechinASD,eveninthecontextofthemany
otherdisablingbehaviorsassociatedwiththedisorder.Thepresentstudyalsoaddstoa
bodyofliteratureonlistenersensitivitytosocial-indexicalaspectsofspeech,butsuggests
thatsomenegativebiasestowardsspeakersmayemergefromthemodelingeffortthatis
neededforlistenerstoengageinsuccessfulcommunicationwithanunfamiliarspeaker.
Classification and evaluation of child speech
38
REFERENCES
AmericanPsychiatricAssociation.(2013).Diagnosticandstatisticalmanualofmental
disorders:DSM-5.Washington,D.C:AmericanPsychiatricAssociation.
Baltaxe,C.A.,&Guthrie,D.(1987).Theuseofprimarysentencestressbynormal,aphasic,
andautisticchildren.JournalofAutismandDevelopmentalDisorders,17,255-271.
Boersma,P.&Weenink,D.(2009).Praat,version5.1.16.Availableathttp://www.praat.org.
Bonneh,Y.S.,Levanon,Y.,Dean-Pardon,O.,Lossos,L.,&Adini,Y.(2011).Abnormalspeech
spectrumandincreasedpitchvariabilityinyoungautisticchildren.FrontiersinHuman
Neuroscience,4,2347.
Clopper,C.G.,&Bradlow,A.R.(2009).FreeclassificationofAmericanEnglishdialectsby
nativeandnon-nativelisteners.JournalofPhonetics,37,436-451.
Clopper,C.G.,&Pisoni,D.B.(2004).Someacousticcuesfortheperceptualcategorization
ofAmericanEnglishregionaldialects.JournalofPhonetics,32,111-140.
Depape,A.M.,Chen,A.,Hall,G.B.,&Trainor,L.J.(2012).Useofprosodyandinformation
structureinhighfunctioningadultswithautisminrelationtolanguageabilitiy.
FrontiersinPsychology,3,1664-1678.
Diehl,J.J.,Watson,D.,Bennetto,L.,McDonough,J.,&Gunlogson,C.(2009).Anacoustic
analysisofprosodyinhigh-functioningautism.AppliedPsycholinguistics,30,385-404.
Diehl,J.J.,&Paul,R.(2012).Acousticdifferencesintheimitationofprosodicpatternsin
childrenwithautismspectrumdisorders.ResearchinAutismSpectrumDisorders,6,
123-134.
Dunn,L.,&Dunn,D.(2007).PeabodyPictureVocabularyTest,FourthEdition(PPVT-4).
Minneapolis,MN:NCSPearson.
Classification and evaluation of child speech
39
Frith,U.(1991).Translationandannotationof“Autisticpsychopathy”inchildhood,byH.
Asperger.InU.Frith(Ed.),AutismandAspergersyndrome.Cambridge,UK:Cambridge
UniversityPress.
Garellek,M.&Keating,P.(2011).Theacousticconsequencesofphonationand
toneinteractionsinJalapaMazatec.JournaloftheInternationalPhoneticAssociation,41,
185-205.
Gertner,B.L.,Rice,M.L.,&Hadley,P.A.(1994).Influenceofcommunicativecompetenceon
peerpreferencesinapreschoolclassroom.JournalofSpeech,LanguageandHearing
Research,37,913-923.
Grossman,R.B.,Bemis,R.H.,PlesaSkwerer,D.,&Tager-Flusberg,H.(2010).Lexicaland
affectiveprosodyinchildrenwithhigh-functioningautism.JournalofSpeech,Language
andHearingResearch,53,778.
Kent,R.D.,Weismer,G.,Kent,J.F.,&Rosenbek,J.C.(1989).Towardphoneticintelligibility
testingindysarthria.JournalofSpeechandHearingDisorders,54,482-499.
Klin,A.,Sparrow,S.,Marans,W.D.,Carter,A.,&Volkmar,F.R.(2000).Assessmentissuesin
childrenandadolescentswithAspergerSyndrome.InA.Klin,F.R.Volmar,&S.S.
Sparrow(Eds.),Aspergersyndrome,(pp.309-339).NewYork:GuilfordPress.
Lass,N.J.,Almerino,C.A.,Jordan,L.F.,&Walsh,J.M.(1980).Theeffectoffilteredspeechon
speakerraceandsexidentifications.JournalofPhonetics,8,101-112.
Lindblom,B.(1990).Explainingphoneticvariation:asketchoftheH&Htheory.InW.J.
Hardcastle&A.Marchal(Eds.),SpeechProductionandSpeechModelling(pp.403-439).
TheNetherlands:KluwerAcademic.
McCaleb,P.,&Prizant,B.M.(1985).Encodingofnewversusoldinformationbyautistic
children.JournalofspeechandHearingDisorders,50,230.
Classification and evaluation of child speech
40
Moore,B.C.J.(1994).Anintroductiontothepsychologyofhearing,3rdEdition.London:
AcademicPress.
Munson,B.(2007).Theacousticcorrelatesofperceivedmasculinity,perceivedfemininity,
andperceivedsexualorientation.LanguageandSpeech,50,125-142.
Munson,B.,McDonald,E.C.,DeBoe,N.L.,&White,A.R.(2006).Theacousticandperceptual
basesofjudgmentsofwomenandmen'ssexualorientationfromreadspeech.Journalof
Phonetics,34,202-240.
Nadig,A.,&Shaw,H.(2012).Acousticandperceptualmeasurementofexpressiveprosody
inhigh-functioningautism:increasedpitchrangeandwhatitmeanstolisteners.
JournalofAutismandDevelopmentalDisorders,42,499-511.
Paul,R.,Augustyn,A.,Klin,A.,&Volkmar,F.R.(2005).Perceptionandproductionof
prosodybyspeakerswithautismspectrumdisorders.JournalofAutismand
DevelopmentalDisorders,35,205-20.
Paul,R.,Bianchi,N.,Augustyn,A.,Klin,A.,&Volkmar,F.R.(2008).Productionofsyllable
stressinspeakerswithautismspectrumdisorders.ResearchinAutismSpectrum
Disorders,2,110-124.
Peppé,S.,&McCann,J.(2003).Assessingintonationandprosodyinchildrenwithatypical
languagedevelopment:ThePEPS-Ctestandtherevisedversion.Clinical
LinguisticsandPhonetics,17,345–354.
Peppé,S.,McCann,J.,Gibbon,F.,O'Hare,A.,&Rutherford,M.(2007).Receptiveand
expressiveprosodicabilityinchildrenwithhigh-functionningautism.JournalofSpeech,
Language,andHearingResearch,50,1015-1028.
Perkins,M.(2007).Pragmaticimpairment.CambridgeUniversityPress.
Port,R.F.(2003).Meterandspeech.JournalofPhonetics,31,599–611.
Classification and evaluation of child speech
41
Purnell,T.,Idsardi,W.,&Baugh,J.(1999).Perceptualandphoneticexperimentson
AmericanEnglishdialectidentification.JournalofLanguageandSocialPsychology,18,
10-30.
Redford,M.A.(2013).Acomparativeanalysisofpausinginchildandadultstorytelling.
AppliedPsycholinguistics,34,569-589.
Redford,M.A.(2014).Theperceivedclarityofchildren'sspeechvariesasafunctionof
theirdefaultarticulationrate.JournaloftheAcousticalSocietyofAmerica,135,2952-
2963.
Redford,M.A.&Oh,G.(2017).Therepresentationandexecutionofarticulatorytimingin
firstandsecondlanguageacquisition.JournalofPhonetics.doi:
0.1016/j.wocn.2017.01.004.
Rice,M.L.,Sell,M.A.,&Hadley,P.A.(1991).Socialinteractionsofspeech,andlanguage-
impairedchildren.JournalofSpeech,LanguageandHearingResearch,34,1299.
Schopler,E.,VanBourgodien,M.E.,Wellman,G.J.,&Love,S.R.(2010).ChildhoodAutism
RatingScale,SecondEdition(CARS2).Torrance,CA:WPS.
Sharda,M.,Subhadra,T.P.,Sahay,S.,Nagaraja,C.,Singh,L.,Mishra,R.,Singhalc,N.,Erickson,
D.,&Singh,N.C.(2010).Soundsofmelody--Pitchpatternsofspeechinautism.
NeuroscienceLetters,478,42-45.
Shriberg,L.D,Kwiatkowski,J.,Rasmussen,C.,Lof,G.,&Miller,J.(1992).TheProsody-Voice
ScreeningProfile(PVSP):PsychometricDataandReferenceInformationforChildren.
UniversityofWisconsin-Madison,WaismanCenteronMentalRetardationandHuman
Development.PhonologyProjectTechnicalReport1.
Classification and evaluation of child speech
42
Shriberg,L.D.,Paul,R.,Black,L.M.,&vanSanten,J.P.(2011).Thehypothesisofapraxiaof
speechinchildrenwithautismspectrumdisorder.JournalofAutismandDevelopmental
Disorders,41,405-426.
Shriberg,L.D.,Paul,R.,McSweeny,J.L.,Klin,A.,Cohen,D.J.,&Volkmar,F.R.(2001).Speech
andprosodycharacteristicsofadolescentsandadultswithhigh-functioningautismand
Aspergersyndrome.JournalofSpeech,LanguageandHearingResearch,44,1097.
Tilsen,S.,&Johnson,K.(2008).Low-frequencyFourieranalysisofspeechrhythm.Journal
oftheAcousticalSocietyofAmerica,124,EL34–EL39.
VanBezooijen,R.(1995).SocioculturalaspectsofpitchdifferencesbetweenJapaneseand
Dutchwomen.LanguageandSpeech,38,253–265
White,L.,&Mattys,S.L.(2007).Calibratingrhythm:firstlanguageandsecondlanguage
studies.JournalofPhonetics,35,501–522.
Classification and evaluation of child speech
43
TABLE1
CharacteristicsofparticipantswithSPEDeligibilitiesforreceivingspeechandlanguage
servicesintheschoolsunderASD.
Speaker Sex Age Symptoms Lg.Delay PPVTp109asd M 6;4 severe No 106
p110asd M 7;1 minimal Yes 119
p102asd M 7;9 severe Yes 126
p100asd M 7;11 severe Yes 83
p103asd M 8;1 moderate Yes 85
p107asd M 8;1 severe No 107
p108asd M 8;1 moderate Yes 112
p114asd M 8;1 minimal Yes 113
p115asd M 8;10 severe No 86
p101asd F 9;3 moderate No 83
p111asd F 9;4 severe No 80
p118asd M 9;6 severe Yes 104
p119asd M 9;6 severe Yes 78
p106asd M 10;3 severe Yes 94
p112asd M 10;9 minimal Yes 102
p113asd F 11;7 severe Yes 90
p116asd M 11;8 severe Yes 94
Classification and evaluation of child speech
44
TABLE2
Characteristicsofparticipantswithtypicaldevelopment.
Speaker Sex Age PPVTp1032td M 6;4 112
p1037td M 7;2 121
p1061td M 7;7 90
p1073td M 7;11 106
p1011td M 8;0 117
p1083td M 8;0 147
p1045td F 8;1 129
p1072td M 8;2 114
p1016td M 9;0 134
p1058td F 9;2 103
p1075td F 9;4 88
p1026td M 9;5 110
p1027td M 9;7 129
p1006td F 10;0 111
p1010td M 10;1 111
p1001td F 10;10 135
p1022td M 10;10 130
Classification and evaluation of child speech
45
TABLE3
Mean(andstandarddeviation)ofuntransformedandunstandardizedresponsesareshownas
afunctionofspeakergroup(TD=typicallydeveloping).Responsesofdisorderobtainedinthe
filteredspeech,unalteredspeech,andtext-basedjudgmenttaskswereaggregatedacross
samplesandlistenerstorenderaby-speakerperceiveddisorderscore(maximumscore=32
forspeech-basedjudgments;16fortext-basedjudgments).Comprehensionoffilteredspeech
wasratedona5-pointscale(nowordsunderstood=1;allwordsunderstood=5).Listeners
alsoratedhowconfidenttheywereintheirresponses(notatall=1;completelysure=5).
JudgmentTask Responsetype Childrenw/TD Childrenw/ASD
Filteredspeech(Experimental)
Cumulative“disordered” 12.53(7.08) 14.71(6.45)
confidence 3.27(0.31) 3.20(0.25)
Comprehension(Control)
ratingsfrom1to5 2.32(0.64) 2.10(0.50)
confidence 3.74(0.89) 3.61(0.91)
Unalteredspeech(Experimental)
Cumulative“disordered” 7.18(5.27) 18.35(7.25)
confidence 3.67(0.32) 3.47(0.19)
Text(Control)
Cumulative“disordered” 4.29(2.80) 7.24(3.54)
confidence 3.41(0.90) 3.38(0.93)
Classification and evaluation of child speech
46
TABLE4
Summaryofindividuallistener’sresponsestotherequestforfeedbackonwhatspeech
characteristicsinfluencedtheirjudgmentsofdisorderinthefilteredandunalteredspeech
conditions.Theresponsesarecategorizedbymajortheme.
Listnr Articulation SpeechClarity Fluency Accenting
1 mispronoun-ciation
slurringofwords
fluidityand…fluctuation
appropriateflux…toaccent
2 slurredwords,concisepronunciation
choppy,flow,pausingatawkwardtimes
emphasisofaword,enthusiastic
3 clarity,unclearslowerspeech,streamingwords,rhythm
4 steadyrhythm,quickly,slowly,smoothly
stressedtherightsyllables,accents
5 enunciated slower,sortofflow upanddown,placedemphasis
6
distorted“r”and“l”sounds,oddlypronouncedblends
7 mispronouncedwords,sounds speechrate stresspatterns
8 slurred,clearer,easiertounderstand littleslower,faster enthusiasm
9 pausesbetween…phrases
mid-phraseinflections
10 soundeddisjoint emphasizecertainpoints,excitement
11 lisp,“r”sounds speakeffectively draggedtheirsyllablesout
12 muffled,extremelyunclear
13 lisp,pronounciating understand…clearly
14 naturalflow,rhythm,paceofwords tone,clashy
15 clearandprecise,slurred quickly
16 disorganizedtones,emphasizedfornopurpose
Classification and evaluation of child speech
47
TABLE5
Allandonlysignificantpair-wisecorrelationsbetweenacousticmeasures.
SchwaDur.
Artic.Rate
F0Slope
FinalF0
MeanF0
H1–H2
H1–A1
SchwaDur. 1 –.43*
Artic.Rate 1 –.52** –.39*
F0Slope 1 –.41* –.44** –.53**
FinalF0Change 1
MeanF0 1 –.61** –.56**
H1–H2 1 .95**
H1–A1 1
*p<.05;**p<.001
Classification and evaluation of child speech
48
FIGURELEGENDS
Figure1.Listeners’judgmentsofdisorder,absenteffectsoflanguage,areshownasa
functionofspeechconditionandspeakergroup(TD=typicallydevelopingchildren;
ASD=childrenwithASDandprosodicdisorder).
Figure2.Listeners’confidenceintheirjudgmentsofdisorderareshownasafunctionof
speechconditionandspeakergroup(TD=typicallydevelopingchildren;ASD=
childrenwithASDandprosodicdisorder).
Figure3.Listeners’ratingsofachild’slikeability(lowanchor=“Awesome!Lovethiskid.”;
highanchor=“Nah.Wouldn’tlikethiskid.”)shownasafunctionofspeechcondition
andspeakergroup(TD=typicallydevelopingchildren;ASD=childrenwithASDand
prosodicdisorder).
Figure4.Listeners’judgmentsofdisorderbasedonfilteredandunalteredspeechsamples
predictsanothergroup’slikeabilityratingsofthespeakerswhoproducedthose
samples.Lowvaluesalongthey-axisindicategreaterlikeabilitythanhighvalues.
Figure5.Theintersectionoflisteners’ratingsofthearticulation,clearness,andfluencyof
unalteredspeechsamples(=intelligibility)predictsanothergroup’sjudgmentsof
disorderbasedonthosesamespeechsamples.Text-basedjudgmentsofdisorder
arepartialledoutfromthevaluesshownalongthey-axis.
Figure6.Theintersectionoflisteners’ratingsofthearticulation,clearness,andfluencyof
unalteredspeechsamples(=intelligibility)predictsanothergroup’slikeability
ratingsofthespeakerswhoproducedthosesamples.Lowvaluesalongthey-axis
indicategreaterlikeabilitythanhighvalues.
Classification and evaluation of child speech
49
FIGURE1
Classification and evaluation of child speech
50
FIGURE2
Classification and evaluation of child speech
51
FIGURE3
Classification and evaluation of child speech
52
FIGURE4
Classification and evaluation of child speech
53
FIGURE5
Classification and evaluation of child speech
54
FIGURE6