Research And Realization Of Speech Recognition Technology Based On DSP

Speech recognization is one of the basements of the personnal-mechian communication. Speech recognization technology has been improved rapidly from the middle of last century. The speech recognization 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 foreground is expansive.Algorithm of the system, the hardware design and the software design were expanded on. The analytical and process foundation knowledge of the speech signal were introduced first. And go further, the fundamental theory of speech recognition was discussed particularly.Peculiarity and architecture of the DSP TMS320VC5402 was introduced, and the speech recognition system was raised. How to design the circuit based on the TMS320VC5402 was researched, and the consist of the system and how the process of the system works was discussed. Analog and digital convert is performed by TLV320AIC23, training and recognizing of the speech signal are performed by DSP, all the logic control is performed by the CPLD chip EPM7032AE, the result of the recognition will be output finally.Contrapositively, we studied the particularness of the Chinese, the procedure of the software design was introduced. The procedure include of pre-processing, start and end point detecting, feature value selection, pattern matching and the training of the formwork. Start and end point detection is the most critical step, we carried out it with short-time average energy. The feature value used in recognition is short-time average energy. A system we will design should recognize a particular speech of a particular person, and the time consumed should be more less, the time domain characteristic we used only. The kernel algorithm is DTW (Dynamic time warping), the algorithm has been certified to be efficient, and a good result has been achieved.