音频事件识别特征核-毕业论文外文翻译

《音频事件识别特征核-毕业论文外文翻译》由会员上传分享，免费在线阅读，更多相关内容在工程资料-天天文库。

1、南京工程学院毕业设计外文翻译（原文及译文）原文名称：FeaturesandKernelsforAudioEventRecognition课题名称：绝缘缘破坏放电声事件检测算法学生姓名学号208130506指导教师所在系部专业名称2017年3月7日FeaturesandKernelsforAudioEventRecognitionAbstractOneofthemostimportantproblemsinaudioeventdetectionresearchisabsenceofbenchmarkresultsforcomp

2、arisonwithanyproposedmethod・Differentworksconsiderdifferentsetsofeventsanddatasetswhichmakesitdifficulttocomprehensivelyanalyzeanynovelmethodwithanexistingone.Inthispaperweproposetoestablishresultsforaudioeventrecognitionontworecentpublicly-availabledatasets.Inpart

3、icularweuseGaussianMixturemodelbasedfeaturerepresentationandcombinethemwithlinearaswellasnon-linearkernelSupportVectorMachines.IndexTerms:AudioEventDetection,AudioContentAnalysis.1・IntroductionInrecentyearsautomaticcontentanalysisofaudiorecordingshasbeengainingatte

4、ntionamongtheaudioresearchcommunity.Thegoalistodevelopmethodswhichcanautomaticallydetectthepresenceofdifferentkindsofaudioeventsinarecording.Audioeventdetection(AED)researchisdrivenbyitsapplicationinseveralareas.Theseincludeareassuchasmultimediainformationretrieval

5、ormultimediaeventdetection[l]wheretheaudiocomponentcontainsimportantinformationaboutthecontentofthemultimediadata.Thisisparticularlyimportantforsupportingcontent-basedsearchandretrievalofmultimediadataontheweb.Otherapplicationssuchassurveillance[2],wildlifemonitori

6、ng[3][4],contextawaresystems[5][6],healthmonitoringetc.arealsomotivatingaudioeventdetectionresearch.AvarietyofmethodshavebeenproposedforAEDinthelastfewyears.AGMM-HMMstructure,similartothatusedinautomaticspeechrecognitionwaspresentedin[7].Asimpleyeteflectiveapproach

7、isthebagofwordsrepresentation[8][9][10].Thisapproachalsofeaturesinaudioeventdetectioninnoisyenvironments[11].Giventhecomplexityofaudioeventdetection,deepneuralnetworkswhichareknownformodelinghighlynon-linearfunctionshavealsobeenexploredforAED[12][13].Someotherinter

8、estingapproachesinvolveuseofacousticunitdescriptorsforAED[14],spectralexemplarsforAED[15],andmatrixfactorizationofspectralrepresentations[16].Lac