Speech Recognition Speed Up Research Based On MFCC

With the development of computer and electronic technology, more and more mobile terminals provide speech recognition, speech instruction and other related functions, whose main purpose is to facilitate people's lives. However, when used for embedded systems, the embedded speech recognition system performs too slowly. The reason is that there are too many floating-point operations in the speech feature extraction process, but these electronic devices do not have strong floating-point operation ability.Aiming at solving the problem that the embedded speech recognition system performs too slowly, fixed-point computing algorithm and look-up table algorithm are proposed in embedded system to speed up speech recognition based on detailed research of Mel frequency cepstrum parameters extraction steps.Recognition rate is an important index when evaluating speech recognition systems. But speech recognition rate and speed is a pair of conflicting parameters. How to improve recognition speed without affecting the recognition rate is another focus in this thesis. Therefore, this thesis verifies the variation situation of speech recognition system implementation speed and recognition rate at three levels experiments from the accuracy, recognition rate and implementation speed in order to compare floating-point speech recognition systems and fixed-point speech recognition systems based on in-depth study on the MFCC extraction algorithm. The results of experiments conducted the contrast of perform pace before and after the speedup and showed that feasible method in this thesis succeeded in accelerating the recognition speed of the system.