Variable Step-Size Speech Blind Separation Employing Laplacian Normal Mixture Distribution Model

《Variable Step-Size Speech Blind Separation Employing Laplacian Normal Mixture Distribution Model》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、VariableStep-SizeSpeechBlindSeparationEmployingLaplacianNormalMixtureDistributionModelZhenhuaZhi,XueyingZhangJianfenMaDepartmentofInformationEngineeringDepartmentofComputing&SoftwareEngineeringTaiyuanUniversityofTechnologyTaiyuanUniversityofTechnologyTaiyuan,ChinaTaiyuan,Chinazhizhenhua@sina.c

2、om;tyzhangxy@163.commajianfen@eyou.comAbstract—Fortheblindsourceseparationalgorithm,choicesofSimulationsofthealgorithm’sperformancearedescribedinnonlinearfunctionandstep-sizeareveryimportant.Inthissection5.Finally,insection6wegiveourconclusions.paper,weproposeanovelnaturalgradientspeechblindse

3、parationalgorithm.ThemainingredientsaretheuseofaII.NATURALGRADIENTBLINDSOURCEvariablestep-sizetechnologyandanonlinearfunctionbasedonSEPARATIONALGORITHMLaplaciannormalmixturedistribution.SimulationresultsindicatetheproposedmethodensuressteadystateerrorofThegoaloftheBSStechniqueistofindaseparati

4、ngalgorithmandacceleratesconvergencespeedofalgorithm.matrixWthatperformstheinverseoperationofthemixingprocess,assubsequentlyusedintheseparationmodel.Figure1showsablockdiagramofthemodel.I.INTRODUCTIONBlindsourceseparation(BSS)technology[1]isanewresearchfieldinlastdecadeofthe20thcentury,isthecor

5、es(k)x(k)y(k)contentinblindsignalprocessingresearchfield,andisalsoacombinationproductofartificialneuralnetworkandstatisticalAWsignalprocessingandinformationtheory.MostformulationstotheBSStaskassumethatasetofsensorsignalscontainlinearmixturesofseveralsourcesignalsofinterest.Thegoalofblindsource

6、separationistorecovertheoriginalsourcesFigure1.BlockdiagramofthebasicBSSgivenonlytheobservedmixtures.Infigure1,Aisanunknown,fullcolumnrankmixingTTTInthenaturalgradientblindsourceseparationalgorithmmatrix.s[s1,...,sn],x[x1,...,xm],y[y1,...,ym].In[2],nonlinearfunctionandstep-sizearetwoimporta

7、ntfactorsthispaper,wesetmn.whichaffectalgorithmperformance.Inordertoimprovealgorithmperformance,wemustselectappropriatenonlinearLetusconsiderthelinearnetworkofthefigure1whosefunction[3][4]andusevariablestep-sizetechnology[5][6].outputy