Hmm-based Isolated Word Speech Recognition System

Speech recogonition is a kind of technology that is using computer to transfer the voice signal to an associated text or command by identification and understand.The speech recognition technology take the pronunciation signal processing as the research object, involves the linguistics, the computer science, the signal processing, the physiology, the psychology and so on many domains, is the pattern recognition important branch. This technology has the extremely broad application prospect. But at present applies the most successful speech recognition system mostly is based on hides Markov to be possible the husband model recognition system, therefore thoroughly understood and grasps hides Mar to be possible the husband model to be extremely significant.The paper main research speech recognition basic principle and the different recognition algorithm, and emphatically study hidden Markov model (HMM) and based on this model speech recognition algorithm. And has constructed one in this foundation based on hidden Markov the model (HMM) speech recognition system.This system has completed the speech recognition main process, contains the pretreatment, the vertex examination, the characteristic parameter extraction and the pattern matching. Pretreated mainly uses the pre-emphasis, the vertex exami- nation uses the double threshold method, the characteristic parameter extraction has used is the linear prediction analysis method, selected has been beautiful your cepstrum coefficient (MFCC) to take the main characteristic parameter, used hidden Markov model (HMM) completed Chinese pronunciation ; the training and the recognition. The result indicated that, this system recognition rate achieved 90% basic may satisfy the system the design request.