The Design Of Speaker Recognition System And Realization On The DSP

Speaker recognition is a kind of biological certification technology and it makes use of the speech coefficients which represent the speaker's physiological and physical feature to identify speaker. In the biological certification area, speaker recognition widely draws the attention because of its convenience, efficiency and accuracy. Recent years, with the development of DSP technology and the improvement of its performance, people are paying more and more attention to the timely speech processing based one DSP. Speaker recognition system based on DSP has higher precision, faster processing speed, smaller bulk, easier operation, lower cost. It can be used in police, law, aerodrome and many other domains and have broad future.The thesis briefly discusses situation and the main theory of speaker recognition. Feature extraction model and pattern matching model as the key factors of speaker recognition, several models for each part such as Pitch, Linear Prediction Coefficient, Mel Frequency Cepstrum Coefficient and Dynamic Time Warping, Vector Quantization, Hidden Markov Model, Gaussian Mixture Model are discussed in detail. Then designs a embedded speaker recognition system based on DSP, and makes great efforts to enhance recognition rate, reduce time and space consume and realize bootstrap operation.The system is based on TMS320C6713 DSK. It regards Mel frequency cepstrum coefficient as the characteristic parameter, adopts the Gaussian mixture model. In order to realize bootstrap operation, the characteristic parameter is stored in flashrom which doesn't lose data when electricity is cut off. We design function to choose training or recognition or depend to train or recognize which person, and to delete, add, replace and renews person. The LEDS belong to TMS320C6713 DSK can show the processing steps and result of the system. In view of the recognition rate, time and space consume , the performance of DSP, the system is designed to recognize ten persons. After testing, The recognition accuracy of this system is over 98%, and the recognition time is about 3 seconds.Compared with other systems, this system has a higher practical value, because its method has been improved which ensures its recognition accuracy. It reduces recognition time, realizes bootstrap operation and can be operated easily.