The Research And Design Of The Speech-Controlled System Based On Fixed-Point DSP

Language is the main way of information transmission for human. Comparing to traditional manual operation, speech has the advantage of simpleness, rapidness and flexibility as the medium of information exchange between human and machine. The speech recognition technology which is the important part of the speech signal processing focuses on speech research and relates to many foundational subjects. It makes the real meaning of man-machine communication become a reality.With the development of speech technology and the popularization of intellectualized system, speech recognition and control become practical gradually and have broad prospect in application.Although speech recognition technology has made gratifying achievements, there are still many issues to be resolved in the actual application. This paper focuses on the application of speaker-dependent, isolated-word and small-vocabulary speech recognition technology and designs a embedded speech-controlled system based on fixed-point DSP.Firstly, this paper analyzes the theory of the front-end processing and gives the simulation results of pre-emphasis, hamming window processing, endpoint detection based on spectral entropy and blind signal separation.Secondly, this paper does research on the Mel-Frequency Cepstrum Coefficients based on properties of the human auditory character about its extraction method and improved technology. On the basis of analysis of the contribution from every coefficient, this paper adopts 26-dimensional optimized parameters as the characteristic parameters of the speech and tests the antinoise ability of the optimized parameters through simulative experiment.Thirdly, after analysis and comparison, this paper selects Dynamic Time Warping algorithm as the recognition method of the system and gives the robust method of training of reference template. In view of the improvement of the accuracy and speed of recognition, this paper does research on the improved technology of the Dynamic Time Warping algorithm, gives the computed path of the template threshold and evaluates the effect of every improved technology through simulative experiment.Finally, according to the technical guideline and the functional requirements, this paper gives the scheme and detail of the hardware design and sofetware actualization respectively. The experiment on DSP evaluation board is carried out to validate the performance of the speech-controlled system designed by this paper.The results of the simulative experiment on PC and the performance of the DSP evaluation board both testify the feasibility and validity of the speech-controlled system designed by this paper. Because the speech-controlled system can easily be connected with multiple execution modules to achieve different functions, it has good practical value in application.