摘要
TheperformanceofthetraditionalVoiceActivityDetection(VAD)algorithmsdeclinessharplyinlowerSignal-to-NoiseRatio(SNR)environments.Inthispaper,afeatureweightinglikeli-hoodmethodisproposedfornoise-robustVAD.Thecontributionofdynamicfeaturestolikelihoodscorecanbeincreasedviathemethod,whichimprovesconsequentlythenoiserobustnessofVAD.Divergencebaseddimensionreductionmethodisproposedforsavingcomputation,whichreducesthesefeaturedimensionswithsmallerdivergencevalueatthecostofdegradingtheperformancealittle.ExperimentalresultsonAuroraIIdatabaseshowthatthedetectionperformanceinnoiseenvironmentscanremarkablybeimprovedbytheproposedmethodwhenthemodeltrainedincleandataisusedtodetectspeechendpoints.Usingweightinglikelihoodonthedimension-reducedfeaturesobtainscom-parable,evenbetter,performancecomparedtooriginalfull-dimensionalfeature.
出版日期
2008年03月13日(中国期刊网平台首次上网日期,不代表论文的发表时间)