Design Of Continuous Speech Recognition System Based On HMM

Speech recognition is a technology that allows the machine to "understand" the human language,and to translate the human speech signal into the corresponding text or command.Speech recognition is a cross subject involving a wide range,with acoustics,linguistics,artificial intelligence,neural psychology,digital signal processing,computer science and other disciplines are closely linked.The application of speech recognition is very extensive,such as voice typewriter,database retrieval,car navigation and so on.This paper focuses on speech recognition,introduced three basic methods of speech recognition,channel model and phonetic knowledge,template matching and artificial neural network.There are three common methods in template matching method: dynamic time warping(DTW),hidden markov(HMM)theory,vector quantization(VQ).In this paper,the speech recognition based on HMM method,most of the large vocabulary,continuous,non independent speech recognition system are based on HMM model.This paper is based on HTK platform,based on HMM model parameters,MFCC is selected as the characteristic parameter,Viterbi algorithm as a basis for identifying,the design of medium vocabulary,speaker independent,continuous speech recognition system is realized.In this design,200 training sentences,100 test sentences,test sentence correctly identified 87,sentence recognition rate is 87%,there are 800 words in test sentences,correctly identified 771 words,word recognition rate is 96.38%.This paper through the comparison monophone construction,triphone modeling and decision tree bound triphone modeling.Finally,it is concluded that decision tree binding of HMM recognition rate is the highest,based on context dependent HMM model has strong anti noise ability and stability.