we give some modifies and demonstrations to some theories in this thesis, as the following:Firstly, the speech signal preprocessing and feature extraction problem have been discussed. We extracted three key features for speech recogm'tion-LPC ,LPCC and MFCC.Secondly, we focused on the discussion on the application of general methods in speech recognition, including the DTW and HMM. What's more, some problem such as scaling, and parameter smoothing have also been discussed.Thirdly, we give some modifies to the general HMM, including explicit state-duration and Stochastic Segment Model(SM), and discussed the training and recognition algorithms when SM applied to speech recognition problem. |