自然口语语音识别研究概况

,,,,21,,;,,,,,,,,21,中国科学院声学研究所副所长本专辑特邀主编杜利民冯俊兰杜利民(中科院声学所语音交互信息技术研究中心)[]概要介绍了自然口语语音识别的研究现状和面临的主要问题,对自然口语语音的特点以及目前正在使用或探索中的算法理论进行较深入地分析和总结:自然口语语音语音识别1,,,40,,,,,90%,,IBMVIAVOICE!,,,,,,,,,,,,,100%,,,,,,∀3∀1999年第9期∀∀;,,,[1]~[3],(flexibility),,,2(spontaneousspeech),!!!,!,,!,,,,,,(filledpauses),[er][um][oh],,[4],,(HMM),HMM,Ngrams,,,,,(),Ngrams,,,331,,,,,,,,,,,,∀4∀∀∀电子科技导报RoseHMIHPY,15[4],,,HMM,,15,15,,,,,,[4],HMIHPY,AT&T,,,,,,,,,32(wordspotting),,,,,,,,,,,,(filler),HMM,,,!,,,,,(downtop),(topdown),[5],(lattice),,(),,,,(subwordmodels),,,,(corpus),[6],,∀5∀1999年第9期∀∀[7],,Viterbi(score),,,Viterbi[8],,,Down-top,,topdown,(context)(triphone),(phone),20,95%,,,,,33,,,(keywordverification)(utteranceverification)LR(O)=pk(O|H0)pk(O|H1),Ok,LR(O),LR(O),O,OkH0(targethypothesis),Ok,H1(antitargethypothesis),Ok,p(O|H0)p(O|H1),(targetmodels)(antitargetmodels),,(MCE)(GPD)[9],,,,,[10,11],[12],[12],,34,,,,,,,,,,,,,[4][8],[13],,,,∀6∀∀∀电子科技导报[14],,,,,,,,,,,4,,,,,,,,,1Rose,R.&Hofstetter,E.Techniquesforrobustwordspottingincontinuousspeechmessages.inproceedingsoftheEuropeanConferenceonSpeechCommunicationandTechnology,pp1183~1186.2Sankar,A.&Lee,C.H.Robustspeechrecognitionbasedonstochasticmatching.ICASSP95I,pp281~284.3Rahim,M..,Juang,B.H,Chou,W.&BuhrkeE.Signalconditioningtechniquesforrobustspeechrecognition.IEEEsignalProcessingLetters,3,107~109.4Rose.ModelingDisfluencyandBackgroundEventsInASRForANaturalLanguageUnderstandingTask.ICASSP99I,pp709~712.5D.A.James,S.J.Young.AFastLattice-basedApproachToVocabularyIndependentWordspotting.ICASSP94,ppI377~380.6RachidaElMeliani,DouglasO#Shaughnessy.Accuratekeywordspottingusingstrictlylexicalfillers.ICASSP97II,pp907~910.7S.dharanipragadaS.Roukos.Afastvocabularyindependentalogrithmforspottingwordsinspeech.ICASSP98I,pp233~236.8TatsuyaKawahara,NorihideKitaoka,ShujiDoshita.ConneptBasedPhasespottingapproachforspontaneousspeechunderstanding.ICSLP96,pp291~294.9W.Chou,B.HJuang,andC.H.Lee.SegmentalGPDtrainingofHMMbasedspeechrecognizer.IEEEInt.Conf.Acoust.SpeechSignalProcessing,1992,pp473~476.10TatsuyaKawahara,ChinHuiLee,BiingHuangJuang.FlexibleSpeechUnderstandingBasedonCombinedKeyphraseDectectionandVerification.IEEEtrans.OnSpeechandAudioProcessingVol.6No.6.NOVEMBER1998,pp558~568.11TakatoshiJITSUHIRO,SatoshiTAKAHASHI,KiyoakiAIKAWA.RejectionofOutofvocabularyWordsUsingPhonemeConfidenceLikelihood.ICASSP98,pp217~220.12E.Lleida,R.C.Rose.EfficientDecodingandTrainingProcedureForUtteranceVerificationInContinuousRecognition.ICASSP96,pp507~510.13R.C.Rose,H.Yao.G.RicardiandJ.Wright.IntegrationofUtteranceVerificationwithStatisticalLanguageModelingAndSpokenLanguageUnderstanding.ICASSP98I,pp237~240.14RameshRangarajanSarukkai,DanaH.Ballard.WordSetProbabilityBoostingforImprovedSpontaneousDialogRecognition.IEEEtrans.onSpeechandAudioProcessing,Vol.5.No.5,September1997,pp438~450.:冯俊兰,博士生主要研究领域为语音识别,讲话者识别与人机对话技术杜利民,研究员,博士生导师,副所长,IEEE高级会员,中国电子学会理事,∃电子学报%编委,∃电子科技导报%编委一直从事语音信号与信息处理技术的研究在IEEETrans.∃电子学报%∃声学学报%等刊物和国际国内学术会议发表研究论文50余篇领导交互信息系统实验室,系统开展汉语语音交互信息系统的关键算法和系统技术的研究∀7∀1999年第9期∀∀