资源描述：
《声纹识别在微信中的模式匹配的研究》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、timewarpingtechnique(DTW),ArtificialNeuralNetwork(ANN),hiddenMarkovmodel(HMM),GaussianMixtureModel(GMM),etc.BecausetheprecisionofDTWishardtoaim,whichleadtoalowrecognitionrate,whileANNneedaverylongtimefortraining,HMMneedalargeamountofcalculation.Therefore,inordertoi
2、mprovetherecognitionaccuracyandefficiency,thispaperselectsthecurrentmainstreamtechnologyoftextindependentspeakerrecognitionofGaussmixturemodel(GaussianMixtureModel,GMM)asmodelingmethod.ThroughthecombinationofthediscreteGMM,withmeanandcovariancematrixtoexpresstheGau
3、ssfunction,soastogettheGMM.BecausetheGaussmixturemodelGMMhasbetterfittingcharacteristicsofdistributionofacousticfeatures,theGMMmethodbasedonthemaximumlikelihooddecisionhasbecomethemainstreammethodofspeakerrecognitionsystem.ItistheGaussprobabilitydensityfunctionofth
4、eextension,thedensitydistributionwhichisabletosimulatevariousshapes.ThetrainingphaseuseEMalgorithmtofindthesetofparameters,andthepatternrecognitionisrealizedbyMAPcriterion.LBGalgorithmisintroducedtocalculatetheinitialvaluesoftheparameters,andthedesignofthecombinedt
5、hresholddecisionisbasedon3methods.Thepaperwitnesseseffectexperimentsofdifferentcharacteristicparametersrespectively,theinitialpointandthethresholdofrecognitionperformance.,TheresultsshowthattheGMMmodelwithmeanvectorandcovariancematrixofthemodelisbetter,whenGaussmix
6、ednumberis32recognitionratereachedthehighest.LBGalgorithmwithhighcompressionratioandlowdistortion,moretogetagoodrecognitioneffect.Combinedthresholddecisioncanreducethefalsepositiverateandfalsealarmrate,andimprovetheefficiencyofidentification.Keywords:VoiceprintReco
7、gnition;Patternmatching;LBG;wechat;GMMIV万方数据目录摘要....................................................................................................................................IAbstract............................................................................
8、...............................................III第一章绪论..........................................................................................
