In this paper, we describe a new method to do speech recognition based on Dynamic HMM architecture. Pitch values are treated as hidden layer and used to modify the parameter of observation probability functions. The results show that the new model achieves approximately 10 percent relative error reductions both in base-syllable recognition task and tonal syllable recognition task. The new method can be used compatibly with conventional HMM based EM training algorithm and Viterbi decoding algorithm.