文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > Statistical Modeling for Unit Selection in Speech Synthesis. In 42nd Meeting of the Associa
侵权投诉

Statistical Modeling for Unit Selection in Speech Synthesis. In 42nd Meeting of the Associa

Traditional concatenative speech synthesis systems use a number of heuristics to define the target and concatenation costs, essential for the design of the unit selection component. In contrast to these approaches, we introduce a general statistical modeli

StatisticalModelingforUnitSelectioninSpeechSynthesis

CyrilAllauzenandMehryarMohriandMichaelRiley

AT&TLabs–Research

180ParkAvenue,FlorhamPark,NJ07932,USA

{allauzen,mohri,riley}@research.att.com

Abstract

Traditionalconcatenativespeechsynthesissystemsuseanumberofheuristicstode nethetargetandconcatenationcosts,essentialforthedesignoftheunitselectioncomponent.Incontrasttotheseap-proaches,weintroduceageneralstatisticalmodel-ingframeworkforunitselectioninspiredbyauto-maticspeechrecognition.Givenappropriatedata,techniquesbasedonthatframeworkcanresultinamoreaccurateunitselection,therebyimprovingthegeneralqualityofaspeechsynthesizer.Theycanalsoleadtoamoremodularandasubstantiallymoreef cientsystem.

Wepresentanewunitselectionsystembasedonstatisticalmodeling.Toovercometheoriginalab-senceofdata,weuseanexistinghigh-qualityunitselectionsystemtogenerateacorpusofunitse-quences.Weshowthattheconcatenationcostcanbeaccuratelyestimatedfromthiscorpususingasta-tisticaln-gramlanguagemodeloverunits.Weusedweightedautomataandtransducersfortherepre-sentationofthecomponentsofthesystemandde-signedanewandmoreef cientcompositionalgo-rithmmakinguseofstringpotentialsfortheircom-bination.Theresultingstatisticalunitselectionisshowntobeabout2.6timesfasterthanthelastre-leaseoftheAT&TNaturalVoicesProductwhilepreservingthesamequality,andoffersmuch ex-ibilityfortheuseandintegrationofnewandmorecomplexcomponents.

1Motivation

Aconcatenativespeechsynthesissystem(HuntandBlack,1996;Beutnageletal.,1999a)consistsofthreecomponents.The rstcomponent,thetext-analysisfrontend,takestextasinputandoutputsasequenceoffeaturevectorsthatcharacterizetheacousticsignaltosynthesize.The rstelementofeachofthesevectorsisthepredictedphoneorhalf-phone;otherelementsarefeaturessuchasthepho-neticcontext,acousticfeatures(e.g.,pitch,dura-tion),orprosodicfeatures.

第1页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email