语音识别技术

11111.2.3.DTW4.HMM5.6.•••••1234•(HMM)(ANN)(500)HMM(100-500)(VQ)(DTW)(10-100)•10------•SonyPSP—TALKMANDTW•DTW----•••DTW••IJ{}12,,,IXxxx=L{}12,,,JYyyy=LDTW•()()(){}()()()()()()1,2,,,nCcccNcninjninjn==LDTWDTW•DTWn()()()()1,,minnjnNCnnyWDijW==∑∑Ninn=1dxnWDTW•C•()()()()1,1−≥−≥njnjnini()()()()JNjINiji====,,111()()()()11,11≤−−≤−−njnjnini()()MMnjni,−DTW•DTW•ij•DTWa(),gij()()()()(){}'''',,,min,,ijnijijgijgijdxyW→=+DTW()()()()11111,1,12,ijgdxy===()()()0,,Re,,,Reijggijijg∈=∞∉()()()(){()()()()()()},min1,,1;1,1,2;,1,3;ijnijnijngijgijdxyWgijdxyWgijdxyW=−+−−+−+DTWIJIJ11()()()131,22nnn===DTW•DTW•DTWDTW••DTW••DTW•DTW•DTW{}{}12111121221222,,,,,,,TTXxxxXxxx==LL?ε()12,dXXYNDTW()()()()()()()()()()()(){}1212DTW,,,,,,12,,,,1,2,,yyyykikjkTyTijijiTjTyxxYyyykT=+==LLL1122••SIDTW•DTWDTW•DTWDTWnmatlabDTW•()DTWHMM•--•DTWmatlabDTW•”abcde“wav(8k8)•DTWmatlabDTW•MFCCMFCC()()21kkikikdnicnii=−=−=+∑∑matlabDTW•MFCC•matlabDTWmatlabDTWmatlabDTWmatlabDTW•---matlabDTW•-(11k16)•matlabDTW•”abc“”de“•distamatlabDTW•b•c•dmatlabDTW•e•matlabDTW•----HMM•10•HMMHMMHMMANN{HMMViterbiλλλ%NYHMMHMM•HMMHMMHMM•HMMHMMHMM••••MarkovHMM•”abcde“wav(11k8)MFCCMakov1243HMM(1)(2)(3)(4)(1)(2)(3)(4)(1)(2)DTW(3)VQ(4)HMM(5)ANN--(1)(2)(3)--(1)(2)(3)(4)(1)VQ(2)/HMM(3)1122NNL--(1)(2)--(3)(4)(1)PCA(2)ANN(3)GMM--•:PP267:1-25TTHHAANNKKSS