基于改进矢量量化k-均值lbg算法的语音识别外文文献翻译

《基于改进矢量量化k-均值lbg算法的语音识别外文文献翻译》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、毕业设计（论文）外文翻译外文题目：SpeechRecognitionUsingVectorQuantizationthroughModifiedK-meansLBGAlgorithm译文题目：基于改进矢量量化K-均值LBG算法的语音识别文献出处:《ComputerEngineeringandTntelligentSystems》，2012,7(3)外文作者：BalwantA.Sonkdinble,DharmpalDoye字数统计：英文2389单词，13087字符；中文3968汉字外文文献:SpeechRecognitionUsingVectorQuantization

2、throughModifiedK-meansLBGAlgorithmAbstractIntheVectorQuantization,themaintaskistogenerateagoodcodebook.Thedistortionmeasurebetweentheoriginalpatternandthereconstructedpatternshouldbeminimum.Inthispaper,aproposedalgorithmcalledModifiedK・meansLBGalgorithmusedtoobtainagoodcodebook.Thesyste

3、mhasshowngoodperformanceonlimitedvocabularytasks.Keywords:K・meansalgorithm,LBGalgorithm,VectorQuantization,SpeechRecognitionlelntroductionThenaturalwayofcommunicationamonghumanbeingsisthroughspeech.Manyhumanbeingsareexchangingtheinformationthroughmobilephonesaswellasothercommunicationto

4、olsinarealmanner[L・R・Rabineretal.,1993].TheVectorQuantization(VQ)isthefundamentalandmostsuccessfultechniqueusedinspeechcoding,imagecoding,speechrecognition,andspeechsynthesisandspeakerrecognition[S.Furui,1986].Thesetechniquesareappliedfirstlyintheanalysisofspeechwherethemappingoflargeve

5、ctorspaceintoafinitenumberofregionsinthatspace.TheVQtechniquesarecommonlyappliedtodevelopdiscreteorsemi-continuousHMMbasedspeechrecognitionsystem.InVQ,anorderedsetofsignalsamplesorparameterscanbeefficientlycodedbymatchingtheinputvectortoasimilarpatternorcodevector(codeword)inapredefined

6、codebook[[Tzu-ChuenLuetal.,2010].TheVQtechniquesarealsoknownasdataclusteringmethodsinvariousdisciplines.Itisanunsupervisedlearningprocedurewidelyusedinmanyapplications.Thedataclusteringmethodsareclassifiedashardandsoftclusteringmethods.Thesearecentroid-basedparametricclusteringtechnique

7、sbasedonalargeclassofdistortionfunctionsknownasBregmandivergences[ArindamBaneijeeetal.,2005].Inthehardclustering,eachdatapointbelongstoexactlyoneofthepartitionsinobtainingthedisjointpartitioningofthedatawhereaseachdatapointhasacertainprobabilityofbelongingtoeachofthepartitionsi