基于數(shù)字串內(nèi)容識(shí)別的說(shuō)話人驗(yàn)證方法研究.pdf

1、中國(guó)科學(xué)技術(shù)大學(xué)碩士學(xué)位論文基于數(shù)字串內(nèi)容識(shí)別的說(shuō)話人驗(yàn)證方法研究姓名：王晉軍申請(qǐng)學(xué)位級(jí)別：碩士專業(yè)：模式識(shí)別與智能系統(tǒng)指導(dǎo)教師：劉振安20050501AbstractThefocusstudycontentofthispaperisspeakerverificationbasedonrecognizetheserialsofnumberwhichuserspeakItisabranchofautospeakerrecognitionA

2、utospeakerrecognitionisanewtechnologywhichidentifytheuserbyhis(0rher)speechIthaswellapplicationprospectsinmanyfieldsByanalyzingspeechcharacteristicparametersandthebasismethodsofspeakerrecognitionandtheirproblems，wechoose

3、LPCCandMFCCandtheirsdifferencetobethespeechcharacteristicparameters，usingverbalinformationverificationofnumberserialstecognizetext—dependentspeechandvalidatethespeakerBeforepickingupthespeechsignalcharacteristicparameter

4、s，ThevoicesignalisundergoingpretreatmentInthisphase，weshouldfindthesignalsstartpointandendpointandfilterthespeechsilencesegmentinordertoprovideusefulspeechsegmentWechoosetheenergy—frequencyvaluemethodExperimentsshowthatl

5、attercanpartitiontheendpointofnoisespeechbetterInthispaper，weusefullpolemodeltoobtainspeechsignalLPC，thendeduceitsLPCC，andweusetheLPCCdifferencetodescribespeakers仃ackdynamicmovementAlsosinceMFCCrepresenthearingfrequencyn

6、onlinearcharacteristic，weutilizeMFCCtobeanotherspeakrecognitioncharacteristicparametertodistinguishtheinputpasswordsAfterintroducethetheoryofHiddenMarkovModel，wegiveadetailedpresentationofverbalinformationverificationmet

7、hodtheoryanditsstepandstepimplement，Afterthat，theresultsofexperimentsanddiscussionhasbeengivenInfactapplication，thesystemcharacteristicparametersautoadaptationcanimprovetheaccuracyofspeakerverificationsystemSowein訂oduces

8、omecharacteristicparametersautoadaptationtheoryAtthesametime，wediscussedmorecombinemodeofspeechverificationandspeakerverificationKeyWords：speakerverification，HiddenMarkovModel，verbalinformationverification，utteranceverif