摘要
Toimprovetheperformanceofsoundsourcelocalizationbasedondistributedmicrophonearraysinnoisyandreverberantenvironments,asoundsourcelocalizationmethodwasproposed.Thismethodexploitedtheinherentspatialsparsitytoconvertthelocalizationproblemintoasparserecoveryproblembasedonthecompressivesensing(CS)theory.Inthismethodtwo-stepdiscretecosinetransform(DCT)-basedfeatureextractionwasutilizedtocoverbothshort-timeandlong-timepropertiesofthesignalandreducethedimensionsofthesparsemodel.Moreover,anonlinedictionarylearning(DL)methodwasusedtodynamicallyadjustthedictionaryformatchingthechangesofaudiosignals,andthenthesparsesolutioncouldbetterrepresentlocationestimations.Inaddition,weproposedanimprovedapproximatel_0normminimizationalgorithmtoenhancereconstructionperformanceforsparsesignalsinlowsignal-noiseratio(SNR).Theeffectivenessoftheproposedschemeisdemonstratedbysimulationresultswherethelocationsofmultiplesourcescanbeobtainedinthenoisyandreverberantconditions.
出版日期
2017年02月12日(中国期刊网平台首次上网日期,不代表论文的发表时间)