Abstract: A voice recognition system based on a double CPU construction with DSP and MCU is designed in the paper. The MCU is used in the control of the voice recognition system. The DSP is used for realizing the preprocessing, the feature extraction and the pattern identification of the voice recognition system. According to the different situation of each stage of the voice recognition system, the wavelet transform and the neural net are introduced into the improvement of traditional realizing methods, which will reduce noises of the system effectively, decrease unnecessary data greatly, and improve the performance of the voice recognition system. The experimental data verify the feasibility and the validity of the method proposed.