Research On Chinese Speech Recognition System Based On HMM And BP Neural Network

Speech recognition technology has gradually changed our life and work in recent years,it can be used to eliminate the barriers of communication between human beings,enhance the man-machine communication.Until now,the technology of DTW,HMM and neural network can be used to realize the speech recognition of non-specific person isolated words.This paper makes a study of the HMM model and BP neural network algorithm,combined HMM model better time series modeling ability with BP neural network technology powerful classification ability,proposed speech recognition method of HMM_BP neural network model,using MATLAB programming.Analyzed and compared the mixed model with the traditional HMM model and BP neural network model.The experimental results show that the hybrid model can effectively improve the recognition rate of the isolated word recognition.(1)Research and analyze the history of speech recognition and the study at home and abroad.Research the basic principles of speech recognition.(2)Study on the whole process of speech processing,including the acquisition of original speech signal,de-noising,pre-emphasis,framing and endpoint detection,selects MFCC as the characteristic parameter.(3)Introduce the three basic problems of HMM model and its use in speech recognition,analyzes the problems and shortcomings.(4)Study BP neural network model and its application in speech recognition,analyzed the advantages and disadvantages of the algorithm in speech recognition.(5)Designed the hybrid model,effectively solved the traditional model of slow convergence speed of BP neural network,the defects of HMM similar words easy to confuse realizing the recognition of independent Chinese isolated words.