基于二次時(shí)頻分布的語(yǔ)音信號(hào)分類與驗(yàn)證.pdf

1、華北電力大學(xué)（北京）碩士學(xué)位論文基于二次時(shí)頻分布的語(yǔ)音信號(hào)分類與驗(yàn)證姓名：鄭普亮申請(qǐng)學(xué)位級(jí)別：碩士專業(yè)：通信與信息系統(tǒng)指導(dǎo)教師：許剛20060101華jE電力人學(xué)碩士學(xué)位論文璃要語(yǔ)音信號(hào)分搬是語(yǔ)音識(shí)別、曉話人識(shí)別和語(yǔ)音合成的一個(gè)重要基礎(chǔ)過(guò)程，其首要閥題就是選擇臺(tái)適的信號(hào)特征參數(shù)。論文營(yíng)先對(duì)辯頻分撰彝囂露售號(hào)楚理嬡突戇發(fā)震皴了麓要密顳，然基對(duì)短辯簿立葉變換、線性預(yù)測(cè)分析、Gabor變換、小波變換和聞態(tài)分析等語(yǔ)港特征表示技術(shù)以及擻小距離分?jǐn)\

2、器、矢量量化、GMM和HMM等算法模型的原理、應(yīng)用場(chǎng)合和特點(diǎn)徽了深入探討。針對(duì)短時(shí)傅立時(shí)變換和小波變換在表驥語(yǔ)音信號(hào)參數(shù)精細(xì)結(jié)構(gòu)上的不足，本文采焉二次時(shí)頻分布徽為語(yǔ)音信母的特征參數(shù)。為獲褥理憨的分類效聚，采用Nelder—Mead算法對(duì)選用的核函數(shù)的參數(shù)進(jìn)行了優(yōu)化。最后，通過(guò)語(yǔ)音識(shí)別和說(shuō)話人識(shí)別實(shí)驗(yàn)顯示，采用優(yōu)化的核函數(shù)，以二次時(shí)頻分布為特征進(jìn)行語(yǔ)音僚號(hào)分類，霹璦獲褥瀵惑熬效采。關(guān)鍵詞：語(yǔ)音識(shí)別，說(shuō)話人識(shí)別，時(shí)頻分布，最小距離分類器AB

3、S至l弧eTClassificationofspeechsignalsisanimportantbaseofthespeechrecognition，speakerrecognitionandspeechsynthesis，forwhichthechiefissueistoselectthepropercharacteristicsofsignalsFirstlywereviewedtheresearchoftimefrequencya

4、nalysisandspeechsignalprocessingThenthedissertationstudiestheprinciples，applicationsandcharacteristicofthespeechcharacteristicrepresentationtechnologies，suchasShortTimeFourierTransform(STF羊)，LinerPredictionCodingLPC)，Gab

5、or曩8nsfcrmandW8velet1’ransform，andarithmeticmodel，suchasMinimumDistanceClassifierVectorQuantization(VQ)GaussianMixtureModel(GMM)andHiddenMarkovModel(HMM)SinceSTFTand、ⅣjveletTransformcan’trepresentthefinestructureofspeech

6、signal，thethesisusesquadratictime—frequencydistributionasthefeatureofthesignalsToobtainthebetterperformanceofclassification，theselectedkernelsareoptimizedbyNelderMeadarithmeticFinallytheexperimentsofspeechrecognitionands

7、peakerrecognitionshowsthatwecanobtaintheperfectperformanceofclassificationwhenquadratictimefrequencydistributionswithoptimizedkernelareusedasehafacto廷sticsZhengPuliang(Communicationandinformationsystem)DirectedbyProfXuGa