基于改進(jìn)網(wǎng)格搜索法SVM參數(shù)優(yōu)化的說話人識別研究.pdf

1、哈爾濱工程大學(xué)碩士學(xué)位論文基于改進(jìn)網(wǎng)格搜索法SVM參數(shù)優(yōu)化的說話人識別研究姓名：王健峰申請學(xué)位級別：碩士專業(yè)：信號與信息處理指導(dǎo)教師：張磊20120314ABSTRACTAsonebiologicalcerrtificationtechnology，speakerrecognitionhastheadVantagesof10wcost，hi曲securityandconvenienceandhasbeen毗ractedpeople’sa

2、ttentionInallalgorithmsaboutspeakerrecogllition，SupponVectorMachine(SVM)algorithInissuitableforsolVingtheproblemofsmalls鋤pleclassmcationandhasbeenoneofthehottestresearchobjectsInthispaper，aimingattheproblemthatSVMtrainin

3、gdatacouldnotbetoobig；thecharacteristicdimensionreductionalgorithmofPrincipleComponentAnalysis(PCA)ismentionedThealgorithmcanreducethenumberoftraⅫngdataa11dimproVetheemciencyofSVMontheconditionofdecreasingtheeffectonther

4、ecognitionrateofsystemasfaraspossibleWhat’smore，aimingattheproblemthatthetimeofsearcllingtheoptimalVectorinParametersoptimizationalgorithmofSVMistoolong，animproVedalgorithmisproposedinthispaperThealgorithmcandecreasethet

5、imeofpar鋤etersoptimizationontheconditionofensuringthesystemrecognitionrateInthispaper，firstly，thesystemstnlcnlreofspeakerrecognitionisanalyzedAndthenthesystemisseparatedtothetwopaIrtsoffeatureextractionandmodelmatchingan

6、dstudiedbasedonthe如nctionofsystemmodulesThepa九offeatureextractionissmdiedonreducingmedimensionoffeaturesAReranalyzingtheadvantagesofPCAincharacteristicdimensionreduction仔omtheory，thee伍ciencyofthealgorithmisproVedwithsimu

7、lationanalysisThepartofmodelmatcllingisstudiedonPar鋤etersoptimizationalgorithmAndt11enthebasicprincipleofPar鋤etersOptimizationA190ritlmlisdiscussedindetailWhat’smore，theperfomallceofParametersOptimizationAlgoritlunsistes

8、tedandtheeffectofParametersOptimizationisanalyzedAt1ast，aimingattheinstan切neityoftheexistingalgorithms，oneimproVedPar鋤etersOptimizationA190rithmisproposedAndthenthespeakerrecognitionsystembasedontheimproVedParametersOpti

9、mizationAlgorithmofSVMintmspaperisproposedThoughthesimulationanalysis，theemciencyoft11eproposedsystemisproVedInthispaper，thoughthemethodofPCA，characteristicdimensionisreducedandthecomplexityofsystemisdecreasedFromthetwov