The Design And Realization Of Voice Control System Based On DSP

With the development of speech recognition technique and the rapid development of DSP chips in recent years, the research of speech recognition and control system have been more and more attended to.The speech recognition technology has been applied to the industry control, communication, and consumption area broadly.Therefore, the research of the speech recognition is meaningful and the application fields is expansive.The research subject in this paper is the voice control system based on DSP, which is aimed for the non-specific isolated word recognition and in order to achieve voice control. The algorithm of voice recognition system, hardware design and software programming are also detailed in the paper.The domestic and international development of the voice recognition technology is firstly introduced and the background of research along with the significance is elaborated. On the basis of the above, the basic theory of voice recognition is introduced according to the model of voice recognition system, which includes the voice signals'pretreatment and the detection of endpoint. Comparing the traditional method of detection, this paper proposes a double threshold comparative method. The feature is extracted and it is focus on the inverted pedigree of LPC and Mel scale frequency. The voice recognition algorithm DTW and HMM is introduced and analyzed in the paper.Followed by the simple depiction about DSP from TI, it proposes the corresponding design method of voice recognition system referring to the existed mature theory foundation. Convert the digital signals from A / D converter of voice signal to the DSP for processing and recognition,the results of recognition were output to follow-up control circuit realized by CPLD ,The signal processing module, voice acquisition module, expanded memory module, CPLD control module and power module are specially designed in this paper.In the software algorithms, Double Threshold Comparative method is applied in the detection of endpoint and and take the Mel cepstrum coefficient as the speech feature parameter. In addition, the discrete hidden markov model (HMM) is adopted, and Baum-Welth reestimation algorithm,forward-backward procedure and viterbi algorithm is utilized to train and recognize the speech signal.The building circumstance of system is introduced, basing on which, the design and debug of the program is completed and proves the feasibility of algorithm. The optimization of program is also researched.Finally the full text is summarized and it points out the future work and the direction to be researched.