The Research On Embedded Speech Recognition

The Speech Recognition is a Technology, which make the machine change the speech signals into homologous text or order by recognition and comprehend. The speech recognition is a cross-subject, and it is becoming to the key-technology of human-computer interface in information technology. The combination of speech recognition and speech synthesis make man can operate the computer through the speech order without the keyboard. Speech recognition system has gained wide application in recent years due to rapid technology developments especially in software and hardware for digital signal processing, the technology for medium and small vocabulary, speaker- dependent and isolated-word speech recognition has come to the age of maturity, it is now used in fields such household electronic and intellectual toys that require relatively low recognition rate.At present, according to the difference of hard core, the research of Speech Recognition System includes: the Speech Recognition System based PC as its hare core and the embedded Speech Recognition System based special chip. The research in the text is the embedded Speech Recognition System using SPCE061A Microcontroller.Firstly, this paper introduces the development of speech recognition and elucidates the background and significance. Secondly, according to the model structure of speech recognition system, it introduces the fundamental theory of speech-recognition. Thirdly, the article introduces one Microcontroller which fits to speech processing. SPCE061A microchip produced by Sunplus Company and its application in the Speech Recognition System. The endpoint detection method is based on SAE-SAZ(short-time average energy and short-time average zero-crossing rate).The system adopts LPCC (linear-prediction cepstral coefficients)with little account quantity as characteristic parameter. Moreover, the demand of the speaker-independent embedded system, in order to reduce memory capacity and calculation, the characteristic parameter is compressed by vector quantity. The discrete hidden markov model (DHMM) is adopted, Baum-Welth reestimation algorithm, forward-backward procedure and viterbi algorithm is utilized to train and recognize the speech signal. The whole algorithms are emulatedin PC by MATLAB.