LVQ神經(jīng)網(wǎng)絡的研究及其在說話人識別中的應用.pdf

1、中國科學技術(shù)大學碩士學位論文LVQ神經(jīng)網(wǎng)絡的研究及其在說話人識別中的應用姓名：程劍鋒申請學位級別：碩士專業(yè)：模式識別與智能系統(tǒng)指導教師：叢爽20040501AbstractThemostsignificantcharacterofapplyingNeuralnetworktopatternrecognitionliesinthatneuralnetworkcanadaptivelylearntherulesoftheinnerofthe

2、inputpatternandtherelationshipsbetweentheinput＆outputpattern。itpresentsapowerfultoolforsolvingthekindofproblemwhichcannotbedescribedbyexactmathlanguageNeuranetworkhastheabilltiesofselforganizing&selflearning，powerfulclas

3、sifying，androbustwithinadequateinformationIteanbeaclassifterwithalmostperfectpropertiesSpeakerrecognition鑫soneofimportantpartsofspeechsignalprocessinghaswidelyresearchingandapplicationprosperitybatthecharacterofthespeake

4、r’svoiceisaffectedbythefactsincludingtime。environmentandbodystatusetc，itstillhasnotbeendescribedbyanexacttheory貳present，soapplyingneuralnetwork犧speakeridentifyinghasvaluablesignificancetheoreticallyandpractically1氌輕paper

5、whichhasthespeakerrecognitionasexperimentalobjectmainlyresearchesontheProbabilisticNeuralNetwork(PNN)andLearningVectorQuantization(LVQ)neuralnetworksdeeplyandsystemicallyT&mainlyresearchresultsare∞follows：Firstly，thePNNi

6、sstudied，theshortcomingsofthePNNareanalyzed，onthebasisofthem，theimprovementonPNNisproposed，andtheFCMwPNNclassifierisconstructedbasedonFuzzyCmeans，SecondlytheLVQneuralnetworkisresearched，Thebasictheory、algorithmsandnetwor

7、kstructureofLVQneuralnetworkaredeseribedAiming氍theshortcomingsofLVQneuralnetwork。TheLVQneuralnetworkisgeneralizedandtherepresentativeformsa糟comparedLastlyOnthebasisofgeneralizedLVQneuralnetwork，TheEMalgorithm，whichisbase

8、donstatisticallearningtheoryisintroduced；ntoLVQneuralnetworkT醅EMLVQalgorithmsproposed，meanwhile，thecorrespondingEMLVQneuralnetworkisconstructedKeyworks：ProbabillstlcNeuralNetwork；FuzzyCmeans；LearningVectorQuantlzmion；Sof