2013_A new adaptive filtering subband algorithm for two-channel.pdf

《2013_A new adaptive filtering subband algorithm for two-channel.pdf》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、ComputersandElectricalEngineeringxxx(2013)xxx–xxxContentslistsavailableatScienceDirectComputersandElectricalEngineeringjournalhomepage:www.elsevier.com/locate/compelecengAnewadaptivefilteringsubbandalgorithmfortwo-channelqacousticnoisereductionandspeechenhancementMohamedDjendi⇑,R

2、édhaBendoumiaUniversitySaadDahlabofBlida,SignalProcessingandImageLaboratory(LATSI),RoutedeSoumaa,B.P.270,Blida09000,AlgeriaarticleinfoabstractArticlehistory:ThispaperaddressestheproblemofacousticnoisereductionandspeechenhancementbyReceived7March2013adaptivefilteringalgorithms.Mos

3、tspeechenhancementmethodsandalgorithmswhichReceivedinrevisedform26Septemberuseadaptivefilteringstructurearegenerallyexpressedinfullbandform.Oneofthese2013widespreadstructuresistheForwardBlindSourceSeparationStructure(FBSS).ThisFBSSAccepted26September2013structureisoftenusedtosepa

4、ratespeechformnoiseandthereforeenhancethespeechsig-Availableonlinexxxxnalattheprocessingoutput.Inthispaper,weproposeanewsubbandimplementationofthisFBSSstructure.Inordertogivemorerobustnesstotheproposedstructure,weadaptKeywords:thenweapplytothissubbandstructureanewcombinationofcr

5、iteriabasedonthesystemSubbandstructuremismatchandthesmoothingfilteringerrorsminimizations.ThecombinationbetweenthisNoisereductionSNRfactorproposedsubbandstructurewiththisoptimalcriteriaallowstoobtainanewtwo-channelMismatchcriterionsubbandforward(2CSF)algorithmthatimprovestheconve

6、rgencespeedofthecrossadap-OptimalMSEtivefilterswhichareusedtoseparatespeechfromnoise.ObjectivetestsundervariousCepstraldistortionenvironmentsarepresentedshowingthegoodbehavioroftheproposed2CSFalgorithm.Ó2013ElsevierLtd.Allrightsreserved.1.IntroductionTheproblemofenhancingspeechco

7、rruptedbyuncorrelatedadditivenoisehasbeenregularlyconsideredinthepastandisstillanactivefieldofscientificresearch[1,2].Severalmethodshavebeenproposedtoresolvethisproblem.Howeveraspecificclassofthesemethodsisessentiallybasedonthesignal-to-noiseratio(SNR)assessmentthatisusedwithagainf

8、unctiontocorrectthecorruptedspeechsignal[3,4].T