資源描述:
《基于fastica的語(yǔ)音盲源分離方法》由會(huì)員上傳分享,免費(fèi)在線閱讀,更多相關(guān)內(nèi)容在行業(yè)資料-天天文庫(kù)。
1、2015年7月四川大學(xué)學(xué)報(bào)(自然科學(xué)版)July.2015第52卷第4期JournalofSiehuanUniversity(NaturalScienceEdition)V01.52NO.4doi:103969/j.issn.0490—6756.2015.07.021基于FastICA的語(yǔ)音盲源分離方法趙忠華,楊曉梅(1.新疆師范大學(xué)物理與電子工程學(xué)院,烏魯木齊830054;2.新疆財(cái)經(jīng)大學(xué)計(jì)算機(jī)科學(xué)與工程學(xué)院,烏魯木齊830012)摘要:獨(dú)立分量分析(ICA)在處理盲信號(hào)分離中被廣泛使用,但其收斂速度較慢.為此文章重點(diǎn)介紹了一種更為有效的盲源分離方法——一快速獨(dú)立分量分析(FastICA)
2、.文章在介紹了FastICA的基本理論和方法之后,將其應(yīng)用到語(yǔ)音分離中.在采集了三個(gè)實(shí)際的聲音信號(hào)后,將三個(gè)原始信號(hào)進(jìn)行混疊,在matlab仿真環(huán)境下用FastICA方法對(duì)混疊信號(hào)進(jìn)行分離,將分離結(jié)果與原始信號(hào)波形進(jìn)行比對(duì),結(jié)果說(shuō)明該算法具有良好的分離效果.關(guān)鍵詞:盲源分離;語(yǔ)音信號(hào)識(shí)別;FastICA中圖分類號(hào):TN912.34文獻(xiàn)標(biāo)識(shí)碼:A文章編號(hào):0490—6756(2015)04—0830—05BlindseparationofsoundsignalbyusingFastICAZHAOZhong—Hua.YANGXiao—Mei。(1.CollegeofPhysicsandElect
3、ronicEngineering,XinjiangNormalUniversity,Urumqi830054,China;2.CollegeofComputerScienceandEngineering,Xi~iangUniversityofFinanceandEconomics,Urumqi830012,China)Abstract:Theindependentcomponentanalysis(ICA)isausablemethodtodealwithablindsignalsep—arationproblem,butwhichconvergesslowly.Inthispaper,am
4、oreeffectivealgorithmofblindsourceseDarationispresented,whichisthefastindependentcomponentanalysis(FastlCA).ItintroducesthatthebasictheoryandapplicationofFastICAinsoundsignalseparation.Threeactualspeechsignalsarefactitiouslymixed,andwhichareseparatedbyusingtheconventionalFastICA.Theseparatedresults
5、showthattheFastICAhasgoodseparationefficiency.Keywords:Blindsignalseparation;Soundsignalseparation;FastlCAcomponentanalysis(ICA)[一]isausableap1IntrOducti0nproachtoestimateoriginalsignalsfromtheobTheblindsignalseparation(BSS)hasmanyservedones.ICAisastatisticalapproachtosolvingtheapplicationssuchasar
6、rayprocessing,imagepro—BSSproblembyalineartransformationtothere—cessing,speechprocessing,telecommunicationsandmedicalsignalprocessing[卜.Thissepara—cordedsignalsSOthattherecoveredsignalsarestatisticallyindependentofeachother一.Theretioniscarriedoutwithoutknowledgeofthemix—aremanydifferentmeasuresofIC
7、Aestimation.ingparametersandthepropertiesoftheoriginalInthisstudy,weintroduceaveryeffectivesignals.InthefieldoftheBSS,theindependentmethodbyfixed-pointiterationscheme,whichis收稿日期:2015—03—13基金項(xiàng)目:新疆師范大學(xué)優(yōu)秀青年教師