adaptive boosting for automatic speech recognition

《adaptive boosting for automatic speech recognition》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、AdaptiveBoostingforAutomaticSpeechRecognitionbyKhamNguyenSubmittedtotheElectricalandComputerEngineeringDepartment,NortheasternUniversityinpartialfulfillmentoftherequirementsforthedegreeofDoctorOfPhilosophy(PhD)attheNORTHEASTERNUNIVERSITYOctober2015○cNortheasternUni

2、versity2015.Allrightsreserved.ProQuestNumber:3741513AllrightsreservedINFORMATIONTOALLUSERSThequalityofthisreproductionisdependentuponthequalityofthecopysubmitted.Intheunlikelyeventthattheauthordidnotsendacompletemanuscriptandtherearemissingpages,thesewillbenoted.Al

3、so,ifmaterialhadtoberemoved,anotewillindicatethedeletion.ProQuest3741513PublishedbyProQuestLLC(2015).CopyrightoftheDissertationisheldbytheAuthor.Allrightsreserved.ThisworkisprotectedagainstunauthorizedcopyingunderTitle17,UnitedStatesCodeMicroformEdition©ProQuestLLC

4、.ProQuestLLC.789EastEisenhowerParkwayP.O.Box1346AnnArbor,MI48106-13462ApprovedforDissertationRequirementsoftheDoctorofPhilosophyDegreeDr.GileadTadmor....................................................DissertationAdvisorProfessorofElectricalandComputerEngineeringDr

5、.JohnMakhoul.....................................................DissertationSecondaryAdvisorChiefScientistatRaytheonBBNTechnologiesDr.JenniferDy........................................................Member,ThesisCommitteeProfessorofElectricalandComputerEngineerin

6、gSpyrosMatsoukas......................................................Member,ThesisCommitteePrincipleScientistatAmazon.comDr.SheilaHemami....................................................DepartmentChairProfessorofElectricalandComputerEngineeringDr.SaraWadia-Fasce

7、tti...............................................AssociateDeanofGraduateSchoolProfessoratSchoolofEngineering4AbstractTheatomicunitsofmostautomaticspeechrecognition(ASR)systemsarethephonemes.However,themostwidelyusedfeaturesinASRareperceptuallinearprediction(PLP)an

8、dmel-frequencycepstralcoefficients(MFCC),whichdonotcarrythephonemein-formationexplicitly.Thediscriminativefeatureswithphonemeinformationhavebeens