In the speech recognition, the matching algorithm is divided into parameteric method, typically HMM(Hidden Markov Model), and non-parameteric method such as DTW (Dynamic Time Warping).
This thesis proposes and designs a new model for the isolated word recognition system, which upgrade recognition accuracy.
As the result of experiments the new model, called HDHE (Hybrid DTW and HMM with Error recovery routine), the recogniyion accuracy scores higher than that of DTW and HMM recognizer.