IMPROVING VOICE ACTIVITY DETECTION VIA WEIGHTING LIKELIHOOD AND DIMENSION REDUCTION

在线阅读 下载PDF 导出详情
摘要 TheperformanceofthetraditionalVoiceActivityDetection(VAD)algorithmsdeclinessharplyinlowerSignal-to-NoiseRatio(SNR)environments.Inthispaper,afeatureweightinglikeli-hoodmethodisproposedfornoise-robustVAD.Thecontributionofdynamicfeaturestolikelihoodscorecanbeincreasedviathemethod,whichimprovesconsequentlythenoiserobustnessofVAD.Divergencebaseddimensionreductionmethodisproposedforsavingcomputation,whichreducesthesefeaturedimensionswithsmallerdivergencevalueatthecostofdegradingtheperformancealittle.ExperimentalresultsonAuroraIIdatabaseshowthatthedetectionperformanceinnoiseenvironmentscanremarkablybeimprovedbytheproposedmethodwhenthemodeltrainedincleandataisusedtodetectspeechendpoints.Usingweightinglikelihoodonthedimension-reducedfeaturesobtainscom-parable,evenbetter,performancecomparedtooriginalfull-dimensionalfeature.
机构地区 不详
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)
  • 相关文献