中英文献翻译:语音识别speechrecognitionSpeechRecognitionVictorZue,RonCole,&WayneWardMITLaboratoryforComputerScience,Cambridge,Massachusetts,USAOregonGraduateInstituteofScience&Technology,Portland,Oregon,USACarnegieMellonUniversity,Pittsburgh,Pennsylvania,USA1DefiningtheProblemSpeechrecognitionistheprocessofconvertinganacousticsignal,capturedbyamicrophoneoratelephone,toasetofwords.Therecognizedwordscanbethefinalresults,asforapplicationssuchascommands&control,dataentry,anddocumentpreparation.Theycanalsoserveastheinputtofurtherlinguisticprocessinginordertoachievespeechunderstanding,asubjectcoveredinsection.Speechrecognitionsystemscanbecharacterizedbymanyparameters,someofthemoreimportantofwhichareshowninFigure.Anisolated-wordspeechrecognitionsystemrequiresthatthespeakerpausebrieflybetweenwords,whereasacontinuousspeechrecognitionsystemdoesnot.Spontaneous,orextemporaneouslygenerated,speechcontainsdisfluencies,andismuchmoredifficulttorecognizethanspeechreadfromscript.Somesystemsrequirespeakerenrollment---ausermustprovidesamplesofhisorherspeechbeforeusingthem,whereasothersystemsaresaidtobespeaker-independent,inthatnoenrollmentisnecessary.Someoftheotherparametersdependonthespecifictask.Recognitionisgenerallymoredifficultwhenvocabulariesarelargeorhavemanysimilar-soundingwords.Whenspeechisproducedinasequenceofwords,languagemodelsorartificialgrammarsareusedtorestrictthecombinationofwords.Thesimplestlanguagemodelcanbespecifiedasafinite-statenetwork,wherethepermissiblewordsfollowingeachwordaregivenexplicitly.Moregenerallanguagemodelsapproximatingnaturallanguagearespecifiedintermsofacontext-sensitivegrammar.1Onepopularmeasureofthedifficultyofthetask,combiningthevocabularysizeandthelanguagemodel,isperplexity,looselydefinedasthegeometricmeanofthenumberofwordsthatcanfollowawordafterthelanguagemodelhasbeenapplied(seesectionforadiscussionoflanguagemodelingingeneralandperplexityinparticular).Finally,therearesomeexternalparametersthatcanaffectspeechrecognitionsystemperformance,includingthecharacteristicsoftheenvironmentalnoiseandthetypeandtheplacementofthemicrophone.ParametersRangeSpeakingModeIsolatedwordstocontinuousspeechSpeakingStyleReadspeechtospontaneousspeechEnrollmentSpeaker-dependenttoSpeaker-independentVocabularySmall(<20words)tolarge(>20,000words)LanguageModelFinite-statetocontext-sensitivePerplexitySmall(<10)tolarge(>100)SNRHigh(>30dB)tolaw(<10dB)TransducerVoice-cancellingmicrophonetotelephoneTable:TypicalparametersusedtocharacterizethecapabilityofspeechrecognitionsystemsSpeechrecognitionisadifficultproblem,largelybecauseofthemanysourcesofvariabilityassociatedwiththesignal.First,theacousticrealizationsofphonemes,thesmallestsoundunitsofwhichwordsarecomposed,arehighlydependentonthecontextinwhichtheyappear.Thesephoneticvariabilitiesareexemplifiedbytheacousticdifferencesofthephoneme,Atwordboundaries,contextualvariationscanbequitedramatic---makinggasshortagesoundlikegashshortageinAmericanEnglish,anddevoandaresoundlikedevandareinItalian.Second,acousticvariabilitiescanresultfromchangesintheenvironmentaswellasinthepositionandcharacteristicsofthetransducer.Third,within-speakervariabilitiescanresultfromchangesinthespeaker'sphysicalandemotionalstate,speakingrate,orvoicequality.Finally,differencesinsociolinguisticbackground,dialect,andvocaltractsizeandshapecancontributetoacross-speakervariab...