Study Of Methods Of Speech Features Extraction Of Ando Tibetan

The study of speech recognition is to force machine to understand what people say.It is the important research direction of speech signal processing.Speech recognition is a synthesis technology that related to acoustics,phonetics,mputer,information processing,and artificial intelligence.With the development of computer,multimedia,and digital processing technology,the more wishes is given to the development of speech recognition.Tibetan speech recognition has just began and from the wide use of other languages recognition,we can say that Tibetan speech recognition will have wide social benefit and economic benefits.It is an indispensable parts of Tibetan information processing.The research of speech signal of Ando Tibetan is a essential step of Tibetan information processing.This thesis used one-syllable of Ando Tibetan as research object.These are the main contents of this paper.On the one hand,according to the way of speech signal processing, Ando-Tibetan is methodical studied.Starting from digital sampling, noise filter,dowing and framing,e've carefully studied the wave of time domain,the wave of frequency domain and the analysis of sonogram. Algorithm with two restricted values threshold are applied to endpoint detection of Ando Tibetan speech signals.We talked algorithms of speaker-independent and isolated-word speech recognition,including DTW,VQ,HMM,and ANN.Several methods of speech feature parameters extraction,such as Linear Predictive Cepstral Coefficients(LPCC),Mel Frequency Cepstral Coefficients(MFCC) are analyzed.The performance of the recognition algorithms based on these parameters are compared in detail.And we amend the methods of features extraction.Experimental results show the efficiency of the new algorithm with△MFCC.On the other hand,with these parameters and methods, speaker-independent and isolated-word of Ando speech recognition system is designed.We've used the GUI technologies of MATLAB to design the interface.The function of this systems are:playing the speech signal,display the parameters' wave that used while recognizing,such as the wave of time domain,the wave of frequency domain,the spectrogram of speech signals,showing the result of endpoint detection of speech signals and recognizing Ando-Tibetan speech.