A Research On Speech Recognition Technology Based On Embedded

Ever since our human beings could make various kinds of machines and use them, we have been had an ideal that all the machines can understand our language and act according to our orally orders, thereby realizing the linguistic communion between men and the machines. With the unceasing development of science and technology and the appearance of Speech Recognition technology, this ideal of human beings has been realized. The Speech Recognition Technology is a high-technology which converts the speech signals into homologous text or orders with machines by speech recognition and comprehension. The speech recognition is an interdisciplinary area, and it is becoming a key-technology of human-computer interface in information technology. The combination of speech recognition and speech synthesis makes man can operate the computer through the speech order without using the keyboard. The application of speech recognition has become a rising competitive industry of high technology. Today, the products based on speech recognition have already played an important part in the application of human-computer interaction.Under this background, based on the deep study on the speech recognition technology, here we offered a scheme that is suitable to speaker-independent and small-scale vocabulary speech recognition system based on SPCE061A micro-controller. It coupled with Data Acquisition and speech input system can be combined into embedded Data acquisition recording system. It will be applied to job-site equipment, instrument Data Acquisition and data entry work and so on in order to improve work efficiency.In this thesis, firstly, the research and development of speech recognition and its existing problems have been introduced, and the basic theory of speech recognition and its implementation methods have been summarized, laying a good foundation for further study of speech recognition. Based on the deep summary in preprocessing technology of the speech recognition, a detailed analyses and comparison of preprocessing techniques have been done, the advantages of various kind of algorithms which can be used to support the realization of the system have been summarized. Accordingly analyses on the way of feature parameter extraction, the principle of feature parameter extraction has been proposed, and also the theory of the model-training and model-matching has been studied. The Discrete Hidden Markov Model (DHMM) is adopted in the speech model, Baum-Welch re-estimation algorithm, forward-backward procedure and Viterbi algorithm have been utilized to train and recognize the speech signal. A basic system that is suitable to speaker-independent and small-scale vocabulary speech recognition based on the SPCE061A micro-controller has been designed.