audio-indexing for broadcast news abstract

audio-indexing for broadcast news abstract

ID:34459221

大小:136.80 KB

页数:5页

时间:2019-03-06

audio-indexing for broadcast news abstract_第1页
audio-indexing for broadcast news abstract_第2页
audio-indexing for broadcast news abstract_第3页
audio-indexing for broadcast news abstract_第4页
audio-indexing for broadcast news abstract_第5页
资源描述:

《audio-indexing for broadcast news abstract》由会员上传分享,免费在线阅读,更多相关内容在教育资源-天天文库

1、Audio-IndexingForBroadcastNewsSatyaDharanipragada,MartinFranz,SalimRoukosIBMT.J.WatsonResearchCenterP.O.Box218YorktownHeights,NY10598ABSTRACTstancesofcontext-dependentsub-phoneclassesareiden-ti edbygrowingadecisiontreefromtheavailabletrain-InthispaperwedescribetheIBMAudio-IndexingSystemingdataan

2、dspecifyingtheterminalnodesofthetreeaswhichisacombinationofalargevocabularyspeechrecog-therelevantinstancesoftheseclasses.Theacousticfea-nizerandatext-basedinformationretrievalsystem.Ourturevectorsthatcharacterizethetrainingdataatthespeechrecognizerwasusedtoproducethebaselinetran-leavesaremodele

3、dbyamixtureofGaussianpdf's,withscriptsfortheNISTSDR97evaluation.Wereporttheperfor-manceofthesystemontheSDR-97knownitemretrieval"diagonalcovariancematrices.EachleafofthedecisiontaskandonamorepertinentTREC-styleAudio-Indexingtreeismodeledbya1-stateHiddenMarkovModelwithtask.aselfloopandaforwardtra

4、nsition.TheIBMsystemexpressestheoutputdistributionsonthestatetransi-tionsintermsoftherankoftheleafinsteadofinterms1.IntroductionofthefeaturevectorandthemixtureofGaussianpdf'sThegoalofanaudio-indexingsystemistoprovidethemodelingthetrainingdataattheleaf.Therankofacapabilityofsearchingandbrowsingth

5、roughaudiocon-leafisobtainedbycomputingthelog-likelihoodofthetent.Thesystemisformedbyintegratinginforma-acousticvectorusingthemodelateachleaf,andthentionretrievalmethodswithlargevocabularycontinu-rankingtheleavesonthebasisoftheirlog-likelihoods.ousspeechrecognition.Alargevocabularycontinuousspee

6、chrecognitionsystemisusedtotoproducetimeForthisSDRtrackevaluationwe,onpurpose,trainedalignedtranscriptsofthespeech.Informationretrievalourrecognizerondatathatisdi erentfromtheSDRtechniquesarethenemployedontheserecognizedtran-(HUB4)trainingdata.Thisenabledustojudgethescriptstoidentifylocationsint

7、hetextthatarerelevantperformanceofoursystemundermismatchconditionstothesearchrequest.Theselocationswithtimealign-whereonewishestoretrieve,asisoftenthecaseinmentsthenspecifyregionsofthespeechthatarerelevantpractice,spokendocu

当前文档最多预览五页,下载文档查看全文

此文档下载收益归作者所有

当前文档最多预览五页,下载文档查看全文
温馨提示:
1. 部分包含数学公式或PPT动画的文件,查看预览时可能会显示错乱或异常,文件下载后无此问题,请放心下载。
2. 本文档由用户上传,版权归属用户,天天文库负责整理代发布。如果您对本文档版权有争议请及时联系客服。
3. 下载前请仔细阅读文档内容,确认文档内容符合您的需求后进行下载,若出现内容与标题不符可向本站投诉处理。
4. 下载文档时可能由于网络波动等原因无法下载或下载错误,付费完成后未能成功下载的用户请联系客服处理。