A-voice-activated-car-audio-system

592IEEETransactionsonConsumerElectronics,Vol.37,No.3,AUGUST1991AVOICEACTIVATEDCARAUDIOSYSTEMS.Tsurufuji:,H.Ohnishi*,M.lida*,R.Suzuki*,andY.Sumi***Information&CommunicationSystemsResearchCenterSanyoElectricCo.,Ltd.,Japan**AudioProductDivision,TottoriSanyoElectricCo.,Ltd.,JapanABSTRACTWehavedevelopedanewspeechdetectiontechnologynamedEnhancedLevel-AdaptiveSEg-mentationmethod(ELASEmethod),whichcandistinguishdriver’scommandsfromaudiosounds.UsingELASEmethod,wecoulddevelopavoiceactivatedcaraudiosystem.ThehighperformanceoftheELASEmethodisclarifiedthroughspeechrecognitionexperimentswhenthecaraudiosystemisplayinginamovingcar.INTRODUCTIONAsdriverscontinuetodemandcaraudiosystemswithmoresophisticatedperformanceandhighprfeatures,thefunctionshavebecomemorecomplex.Therefore.thenumberofbuttonsinIron1parwlofthecaraudiosystemareincreasing.Yetmorecomplexfunctionsmaycausetrafficaccidentssincetheyrequiremoreofthedriver’sattention.Sowehavebeenhopedtooperatethecaraudiosysteminvoicecommands.Recently,speechrecognitionsystemshavecomeintopracticaluse.[11Insuchconventionalspeechrecognitionsystems.aspokenwordisconvertedintoanelectricalsignalwithamicrophone,andthewordboundaryisdeterminedbymonitoringtheamplitudeoftheelectricalsignals,SOthataspeechrecognitionmaybeachievedthroughspectrumanalysisandrecognition.Therewillbenoproblemsifsuchspeechrecognitionsystemsareusedunderaquietenvironment.Butsomeproblemsremainunderanoisyenvironment,especiallyinamovingcar.Inthispaper,wedescribetheproblems,especially,anoiseenvironment,whenthecaraudiosystemisplayinginamovingcar.Tosolvethisproblem,weproposeanewspeechdetectiontechnology.withwhichwehavedevelopedavoiceactivatedcaraudiosystem.Fig.1AVLewofAVoiceActivatedCarAudioSystemManuscriptreceivedJune7,199100983063/91/020000592$01.001991IEEE-Tsurufuji,etal.:AVoiceActivatedCarAudioSystemAmlysismethodMatchirgmethodVocabuarysizeUiteranceduaticnResponcetimeFEATURESOFSPEECHRECOGNITIONSYSTEM8chamelbardpassfiIterLinearmtching&Dynamictimwarping21words0.2-1.5second0.4second(typ)Wehaveconsideredthefollowingfourrequirementswhendevelopingavoiceactivatedcaraudiosystem;(1)hands-freeoperationbyvoicecommands,(2)acaraudiosystemmustbeindoubleDIN(3)lowercostforcommercialuse,(4)thepracticalusewhenthecaraudiosystemisplayingwherethepeaklevelis80dB.Thespecificationofspeechrecognitionsystemsize.wehavedevelopedisshowninTable1.[2]Table1SpecificationIRecognitiontywSpeakerdependentIisolatedwordThespeechrecognitionsystemisnowmadeofaspeechrecognitionmoduleprintedonasmallboard(70(W)x30(D)x12(H)mm).ThespeechrecognitionmoduleconsistsfromahybridICforanalysis,amemoryICandarecognitionLSI.Usingthecompactspeechrecognitionmodule,wecouldsolvethethreerequirements(hands-freeoperation,sizeandcost).Butthelastrequirement(thepracticalusewhenthecaraudiosystemisplaying)couldn’tbesolvedbytheconventionaltechnology.THEPROBLEMUNDERNOISYENVIRONMENTToobtainthehands-freeoperationbyspeechrecognitionduringhighspeeddriving,therobustnessagainstthenoiseisveryimriortarit.1:11Thesearesomereasonsasfollowswhichmakespeechrecognitionsodifficult;(1)theroadnoisecausedbyhighspeeddriving.(2)thenoisefromengine,(3)thesuddennoisefromoutsideofthecar.Whenthecaraudiosystemisplaying.thespeechrecognitionismoredifficultbythefollowingreason;(4)theaudiosoundfromcaraudiosystem.Whenmanypeop1ehaveconversationsinthe593car,thespeechrecognitionbecomesmoreandmoredifficult,(5)theconversationbetweentheThereasons(1)and(2)have1ittonspeechrecognitionbecausewedirectionalmicrophoneandthedmicrophonetomouthislimitedwiinches.(3)andotherpeople.Wetakecareofreasonsdriverandeinfluenceuseauni-stancefromhinabout3(5)byvoicere-entrymethodand”password”system.Thespeechrecognitionsystemfailstocatchthedriver’svoicecommands.whentherecog-nitionthresholdtoeliminatenoiseissettoohigh.There-entrymethodmeanstosetitlowerandlowerautomaticallyforthesecondandthirdre-entry.Itmakespossibletocatchanyspokencontrolwordstwiceorthreetimes.Buttheabovemethodscouldn’ttakecareofreason(4).Bytheway,somesolutionmethodsinanoisyenvironmentarealreadyknownasfollows;0arobustanalysismethodforthenoise[4],@anoisesubtractiontechnology[51,0aspeechdetectiontechnologyundera0aspecialmatchingmethodforthenoise.Inthespeakerdependentisolatedwordrecog-nitionunderanoisyenvironment,weconsid-erthatthespeechdetectionisveryimportant.noisyenvironment[61,ASPEECHDETECTIONTECHNOLOGYThecaraudiosoundispleasantforthedrivers,butitis”noise”forthespeechrecognitionsystem.Sincetheconventionalboundarydetectionmethodofspokenwordusesafixedthresholdmethod,thebeginningframeofspokenwordisdeterminedealierthanthecorrectbeginningframeinthesoundsofcaraudiosystem.TheboundarydetectionofspokenwordwiththefixedthresholdmethodisshowninFig.2.Fig.2Boundrydetectionwithfixedthreshold594IEEETransactionsonConsumerElectronics,Vol.37,No.3,AUGUST1991Sincetheaudiosoundlevelisalreadyknownforthecaraudiosystem,wedecidedtoutilizetheaudiooutputlevel.Inthisvoiceactivatedcaraudiosystem,aaudioleveldetectionunitisincluded.Thisaudioleveldetectionunitchangesthesignalofaudioamptoanaudiooutputsignal.Theblockdiagramo