汉语连续语音识别系统的研究与实现(3)
发布时间:2021-06-08
发布时间:2021-06-08
语音信号处理相关论文
Abstract
Speechrecognitionisatechnologythatcanextractcharactersymbolsfromthespeechsignalthroughcomputerprocessing.Chinesecontinuousspeech
arerecognitionsomehasbeeninresearchedforalmost60yearsathomeandabroad.Althoughthereachievements
continuous
speechspeechrecognitionresearch,atechnologiescalllotofproblemsneedtobeCallresolved.Existingcommunicatewim
speechrecognitionnotreachthetargetthathumanmachinebynaturalspeech.Largevocabulary,speaker-independentcontinuous
recognitionisstillthedifficultyandkeystoneofspeechrecognition.
continuousspeechThemainstudyofthispaperisthekeytechnologiesofChinese
recognition.Firstly
ofspeechthepaperintroducestheprinciplesofspeechrecognition,thecompositionthenrecognition
thefunctionssystemandandthebasicknowledgeofChineseSpeech.Andintroduceskeytechnologiesofpre-processing、featureextraction、pattern
ofspeechmatchingand
inviewof
follows:post-processingexistedrecognition.Improvedmethodshavebeenproposedareasproblemsintraditionalmethods.Inthispaper,themainworks
1)Amedium-vocabulary,speaker-independentCllinesecontinuousspeech
Visualrecognitionsystemisachievedonapersonalcomputer,usingMicrosoft
onC++,MATLAB,MicrosofiSQLServerandothertools,experimentsaremade
unit,thethesystem.Thesystemchoosestheinitialandfinalas
theDTWasrecognitionMFCCasfeatureparameters,andusingrecognitionmodel.
accuracyofI/F2)The
segmentation
continuoussegmentationhasgreatinfluenceonsystem,currentI/Fmethodsachievehighdegreeofaccuracyinnewmethodnon continuousspeech,butfallinofspeech.Inthispaper,abasedonentropyandtheformantenergy
FF.Chinesevowelsisproposed,whichcanaccuratelysegmentthe
long3)Thespeechrecognitionsystem、ⅣiththetraditionalDynamicTimeWarpingtechnologyhasgreatamountofcalculationandresponse
methodtime.Forthisimproved
andmethodsofDTWmethod
canaleintroduced:animprovedonproblem,twobasedontemplatethresholdexperimentsshowresponsetime。animprovedbasedfeaturevectorthreshold.Theresultsofthatthenewmethodsreducethecalculationtimeandimprovethe
Keywords:SpeechRecognition,EndpointDetection,MelFrequencyCepstralCoefficient
(MFCC),DynamicTimeWarping(Dn聊11