Study On Hybrid Model Of Speech Recognition Based On HMM And PNN

Speech recognition is a comprehensive discipline, which contains linguistic, signal processing and the computer technology. Its broad application prospect and widespread application domain caused the massive scientific researchers to have paid attention and put to it for a long time .Firstly, this article has carried on the analysis domestic and foreign present development situation of speech recognition technology, and introduced the speech recognition elementary theory which concluding composition, classified, preprocessing, feature extraction as well as double threshold detecting method and so on. Introduced the linear prediction cepstrum coefficient and Mel-cepstrum coefficient's principles and extracting process in detail.The principle and application in speech recognition of HMM model and PNN have been studied and analyzed and compared of their respective recognition property advantages and weakness in depth. Combining HMM powerful time domain modeling ability with PNN outstanding classification ability, proposed a hybrid model and algorithm which based on HMM and PNN that use VQ as HMM front-end, after vector quantified the LPCC parameters and MFCC parameters which have been secondary extracted, the quantified results used as input of HMM. On this basis, the states translate accumulated output which generated by Viterbi decoding as the input of PNN, then PNN output the final recognition results.In this paper, HMM and HMM / PNN model has been simulated by Matlab7.0, and the recognition rate which under conditions of noise were compared. Identification data obtained from the simulation can be proved that compared with the traditional HMM, the recognition of the hybrid model has better performance and t anti-noise performance is more superior...