(WORD)外文翻译基于仿生模式识别的非特定人连续语音识别系统其他专业

《(WORD)外文翻译基于仿生模式识别的非特定人连续语音识别系统其他专业》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、中文3276字附录英文原文：ChineseJournalofElectronicsVo1.15,No.3,July2006ASpeaker--IndependentContinuousSpeechRecognitionSystemUsingBiomimeticPatternRecognitionWANGShoujueandQINHong(LaboratoryofArtificialNeuralNetworks，InstituteolSemiconductors，ChineseAcademySciences，Beijing100083，China)Abstract—In

2、speaker-independentspeechrecognition,thedisadvantageofthemostdiffusedtechnology(HMMs，orHiddenMarkovmodels)isnotonlytheneedofmanymoretrainingsamples，butalsolongtraintimerequirement.ThisPaperdescribestheuseofBiomimeticpatternrecognition(BPR)inrecognizingsomemandarincontinuousspeechinaspea

3、ker-independentManner.Aspeechdatabasewasdevelopedforthecourseofstudy．Thevocabularyofthedatabaseconsistsof15Chinesedish’snames,thelengthofeachnameis4Chinesewords．Neuralnetworks(NNs)basedonMulti-weightneuron(MWN)modelareusedtotrainandrecognizethespeechsounds．ThenumberofMWNwasinvestigatedt

4、oachievetheoptimalperformanceoftheNNs-basedBPR.Thissystem,whichisbasedonBPRandcancarryoutrealtimerecognitionreachesarecognitionrateof98.14％forthefirstoptionand99.81％forthefirsttwooptionstothePersonsfromdifferentprovincesofChinaspeakingcommonChinesespeech．Experimentswerealsocarriedontoev

5、aluateContinuousdensityhiddenMarkovmodels(CDHMM),Dynamictimewarping(DTW)andBPRforspeechrecognition．TheExperimentresultsshowthatBPRoutperformsCDHMMandDTWespeciallyinthecasesofsamplesofafinitesize．Keywords—Biomimeticpatternrecognition,Speechrecogniton,HiddenMarkovmodels(HMMs),Dynamictimew

6、arping(DTW)．I．IntroductionThemaingoalofAutomaticspeechrecognition(ASR)istoproduceasystemwhichwillrecognizeaccuratelynormalhumanspeechfromanyspeaker．Therecognitionsystemmaybeclassifiedasspeaker-dependentorspeaker-independent．Thespeakerdependencerequiresthatthesystembepersonallytrainedwit

7、hthespeechofthepersonthatwillbeinvolvedwithitsoperationinordertoachieveahighrecognitionrate．Forapplicationsonthepublicfacilities，ontheotherhand，thesystemmustbecapableofrecognizingthespeechutteredbymanydifferentpeople，withdifferentgender，age，accent,etc.,thespeakerindependencehas