返回
e s e a r c h r e p r o r t i d inew

e s e a r c h r e p r o r t i d inew

ID：34432292

大小：251.01 KB

页数：14页

时间：2019-03-06

e s e a r c h r e p r o r t i d inew_第1页
e s e a r c h r e p r o r t i d inew_第2页
e s e a r c h r e p r o r t i d inew_第3页
e s e a r c h r e p r o r t i d inew_第4页
e s e a r c h r e p r o r t i d inew_第5页
资源描述：

《e s e a r c h r e p r o r t i d inew》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、ORTIDIAPMartigny-Valais-SuisseEPRMulti-StreamSpeechRecognitionabHerveBourlardStephaneDupontbESEARCHChristopheRisRIDIAP{RR96-07IDIAPDecember1996DalleMolleInstituteforPerceptiveArtificialInteligenceP.O.Box592MartignyValaisSwitzerlandphone+41?27?721771

2、1fax+41?27?7217712e-mailsecretariat@idiap.chaIDIAPinternethttp://www.idiap.chbFacultePolytechniquedeMons,BelgiumIDIAPResearchReport96-07Multi-StreamSpeechRecognitionHerveBourlardStephaneDupontChristopheRisDecember1996Abstract.Inthispaper,wediscussanewa

3、utomaticspeechrecognition(ASR)approachbasedonindependentprocessingandrecombinationofseveralfeaturestreams.Inthisframework,itisassumedthatthespeechsignalisrepresentedintermsofmultipleinputstreams,eachinputstreamrepresentingadi erentcharacteristicofthesigna

4、l.Ifthestreamsareentirelysynchronous,theymaybeaccommodatedsimply(astheyusuallyareinstate-of-the-artsystems).However,asdiscussedinthepaper,itmayberequiredtopermitsomedegreeofasynchronybetweenstreams.Thispaperintroducesthebasicframeworkofastatisticalstructu

5、rethatcanaccommodatemultiple(asynchronous)observationstreams(possiblyexhibitingdi erentframerates).Thisapproachwillthenbeappliedtotheparticularcaseofmulti-bandspeechrecognitionandwillbeshowntoyieldsigni cantlybetternoiserobustness.2IDIAP{RR96-071Introduct

6、ionIncurrentautomaticspeechrecognition(ASR)systems,theacousticprocessingmoduletypicallyemploysfeatureextractiontechniquesinwhich20to30millisecondsofspeechisanalyzedoncepercentisecond,leadingtoasequenceofacoustic(feature)vectorsthateachdescribelocalcompone

7、ntsofthespeechsignal.Eachacousticvectoristypicallyasmoothedspectrumorcepstrum.HiddenMarkovModel(HMM)states,whicharetypicallyassociatedwithcontext-dependentphonessuchastriphones,arethencharacterizedbyastationaryprobabilitydensityfunctionoverthespaceofthese

8、acousticvectors.WordsandsentencesarethenassumedtobepiecewisestationaryandrepresentedintermsofasequenceofHMMstates.Instate-of-the-artASRsystems,each10-msspeechsegmentisoftendescribedintermsofseveral(dependentorindepe

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。