Research And Implementation Of Wavelet Transform In Speech Recognition System

Automated Speech Recognition is a technology that it converts speech to text.Speech Recognition correlates with multi-subject.It is joined with acoustics, phonetics, linguistics,digital signal process theory, information theory, computer science and so on .The more deep We reseach ASR ,the more goal people expect .The purpose of ASR is that realizes free communion between people with machine.That is make machine has hearing.It can comprehend meanings of language,it can convert language to significative signal ,it makes machine working according to people will,it makes people breaking away from heavy or dangerous work. The Speech Recognition is forecasted that is next overfly,just as RichardAwetchkenbaum ,PC analyst of IDC said :language is most natural interface.The Speech Recognition has prodigious actual application value,its development, maturate,and practicality will impel development of many industry that include computer,OA, communications, national defence,robot and so on .Now the application of the Speech Recognition includes :the speech input system as a most natual input method ,replacing keyboard input character with dictation,it brings OA and publish industry with revolutionary change;the speech contral system provides people with more safety and convenience control way, especially some specifically environment or especial custom ;the database query system basing on dialogue system provides custom with more natual ,friendly and convenient database searches and query,it can apply in the Bank , the Bourse, civil aviaton and so on;.and besides, the speech recognition can apply in spoken language translate system,CAD,automation identity affirm etc. Because of diversity and complexity of speech signal, actual speech recognition system obtains satisfying capability just on specifically condition . The capability of speech recognition system approximately lies on four diathesis:1. size of distinguish vocabulary and complexity of speech.2. quality of speech signal 3.single addressor or multi- addressor.4. hardware.Because theory can not make a breakthrough?,research work gets tardiness.Through there are some new method,but their applicability is bad .So difficult is : ??●The adaptability of the Speech Recognition system is bad,such as depending on environment;another questions is that it can not accurately respond to input of custom, it is discommodiousness. ●Speech recognition is difficult in the heavy noise ?environment,because diversification of people pronunciation is complex,such as speech becomes high,speech becomes slow and so on, this is Lombard offect ,so we must look for new signal analysis method.??●There are many success of Linguistics, physiology and psychology,but we must study how to apply it in speech recognition . Furthermore it is very important that we use language model , syntax and accidence model in middle or big glossary continuous speech recognition. ???●We are not very understand about people hearing comprehend, information accumulate,study way and neural control theory; secondly,it is a difficult course that we apply it in speech recognition .??●We have many question to solve during the couse that lab demo of the Speech Recognition system converts to merchandise ,such as recognition speed,no identify and key detect technology and so on.?? For the sake of soloving these question , technologist find out various way ,such as adaptself, basing on ear characteristic, analysing and distilling parameter, ANN etc.And we obtain some success.Wavelet analysis is applied in image process and speech process as a signal analysis method, it is partial convert of time and frequency,it can effectively distill imformation from signal.After wavelet transform we can obtain better time exculpation rate in the high frequency field,we can obtain better frequency rate in the signal low frequency field.This characteristic of wavelet adapts to speech signal process specially.Because characteristic of wavelet is similitude to apperceive couse of ea...