资源描述:
《特定词汉语语音识别系统应用研究》由会员上传分享,免费在线阅读,更多相关内容在学术论文-天天文库。
1、广西师范大学硕士学位论文特定词汉语语音识别系统应用研究姓名:叶洪涛申请学位级别:硕士专业:电路与系统指导教师:黄汉明20060101AppliedResearchonSpecificWordChineseSpeechRecognitionSystemGraduatestudent:YeHongtaoAdviser:HuangHanmingGrade:2003Speciality:CircuitandSystemResearchdirection:Patternrecognitionandsignalprocessing
2、AbstractAtpresent,thespeechrecognitiontechnologyhasalreadybeenwidelyappliedindiverseareas,suchas:householdelectronics,intellectualtoys,thedatabasevoiceinquiryofthebusinesssystem,industrialvoicecontrol,autodialingoftelephoneandtelecomsystem,etc.,andwillaffirmativ
3、elybecometheinterfaceofthenextgenerationoperatingsystem.ThispapermakesanattemptontheinvestigationinDTWandHMMwiththeirapplicationstoChinesespeechrecognition.Atfirst,variousaspectsofthefundamentalsofthespeechrecognitionsystemarediscussedindetail.Thenpreprocessinga
4、ndfeatureextractionarefollowed.Afterthatarethewindowing,noisefiltering,endpointdetection.Twoimportantspeechanalysismethodsforspeechrecognition:MFCCandLPCCareintensivelyinvestigatedandcompared.Intheresult,MFCCisselectedasfeatureselectorforspeechsignals.Atthesecon
5、dphase,severalspeechrecognitionalgorithms,includingtheDTWandHMM,arethoroughlydiscussed.WehaveexperimentallyproventheimprovedDTWoutbalancingthefundamentalDTWbyrecognitionefficiency.Also,weproposedanovelseekingmethodforconfirmingreferencepatterns.Themethodisthatfi
6、rstlycomputingtheDTWcumulativedistancesbetweeneachotherofseveralcharacteristicvectorsequencesforeachword,thensummingupthesedistancestoeveryonevectorsequencefromallothervectorsequencesforthesameword,selectingtheminimumfromthesesummeddistancesandthecorrespondentch
7、aracteristicvectorsequenceregardedasthereferencepatternforthatword.Thisnovelseekingmethodhasexperimentallyshownexcellentrecognitionaccuracy.ThenweaddressHMM’sfundamentalsandHMM'sparameterselectionproblems.Thelastbutthemostimportant,bothofDTWandHMMalgorithmsareim
8、plementedonaPC,andthecomparisonsbetweenthesetwoalgorithmforefficiencyandaccuracyarecarriedoutandtheresultsarelistedinthispaper.KeywordsSpecificWordSpeechRecognition,P