基于數(shù)據(jù)融合的語(yǔ)音情感分析與識(shí)別.pdf

1、y79UG0‘分類號(hào)：I型!垡：曼密級(jí)：j蘭丑I_UDC：占皇f三學(xué)號(hào)：0__20425東南大掌碩士學(xué)位論文基于數(shù)據(jù)融合的語(yǔ)音情感分析與識(shí)別研究生姓名：夔麴塹：導(dǎo)師姓名：壑力夔攮申請(qǐng)學(xué)位級(jí)別王堂亟論文提交日期至壘Q!生圭旦皇￡旦學(xué)位授于單位壅直盔堂答辯委員會(huì)主席堇童寶學(xué)科專業(yè)名稱籃呈皇籃星處堡論文答辯日期呈Q墮笙曼旦!l目學(xué)位授予日期呈坌Q曼生主月4目評(píng)閱人剎睦矍鹽2005年3月1日東南大學(xué)碩士學(xué)位論文AbstractSocialper

2、ceptionsandjudgmentsofemotionareubiquitousandaredeemedimportanttohumansurvivalWiththedevelopmentofscience，theprocessingandanalysisofemotionWaSalreadybecameallimlxntantresearchdirectioninthefieldofArtificialIntelligenceIn

3、traditionalspeechs刪processing，thecomputercanonlyrecognizetheinformationaboutsymbol，thehiddeninfmmadourelatedwithemotionandmoodisdroppedwithoutbeennoticedThedevelopmentofspeechemotionalanalysiswillimproveandenhancethecomp

4、uter’Semotionalintelligancetheabilitytecognizeausersaffoctivestates，tobecamemorehumanlike，moreeffectiveandmoreefficientThecontentsinthispaperareincluding：thebuildingandtestingofemotionaldatabasecalculationvalidityestimat

5、ingandclassificationofparamet日semotionalrecogaitinnbased∞datafusionAreviewofthehistoryOfemotiontheoryandthereseat℃hofspeechemotionmin打oducedatfirstinthispaperBasedontheresultscomefromtherese口chwiththephysiologicalpsychol

6、ogysomeclasstllcntionsoftheemotionarelistThebllildingofemotionaldatabaseisoumideredSomedatabasescurrentlybeingusedareOesenbe吐theshortcomingandadvantagefordiffeaentdatabasesarepointedoutBasedOIltheeanalyses，aCMneseemotion

7、alspeechdatableiseoustruetedItincludestheacteddataofwordssentencesandpassagesandsixkindsofemotionsofhappiness，angersurprisesadnessandfearThetestingiscarriedoutusingtheknowledgeof蛔mathematicsandtheaimisdecreasetheintefc∞n

8、ceofsubjectivefactorsPitchformantenergytherateofsix；cchandf1aotaldimensionofspeechwaveandpitchenlwearecalculatedasparametersinspeechemotionalrecognitionThenowmethodofegtlaOtillgmaxilnwnandmininlnnlisintroducedTheoryofthe

9、fuzzyentropyhasbeenusedtomeasmotheuncertaintyofthesevariablesTwokindsofclassifyingmethodtoparametexs躺carriedoutforfusionanalysisAnovelreeo窖丑]iziagmethodofmultidassSVMKNNisproposedbasedmtherelationshipbetweenSVM(SupportVe

10、ctorMachine)andKNNⅨNearestNeighbor)SVMKNNandSynergetienetworkaleadizadforemotionrecognitioninthispapeaTherestdtsthatSVMKNNismoreeffectivelythan‘‘OneAgainstAfi’SVMSynergetienetworkismoreeffectivelythanmodifiedPCA(Principl

11、eComponemAnalysis)areprovedthroughexperimentsRecognizingmethod鋤dfusionalgorithmarecombinedinthispaperThroughcombiningadaptiveweighftlSiORandSVMKNN，mutielassfusionandSyaergetienetworktherateofspeechemotionalrecognitionise

12、nhancedAverageaeceraoyisadvanced446％forfemale，266％formaleInaddition，adaptiveweighfnsimaisprovedmofeeffectivelythanthemethodofmuticlassfusionKeywords：SpeechEmotionReeognitiou，F(xiàn)uzzyEn扛opymuticlassfusionSupportVectorMachine