Speaker recognition is a kind of biological authentication technology which distinguishes speakers' identity by matching the voice distilled beforehand. However, the noise circumstance is an obstacle disturbing this technology walking up to practicality. Concerning the shortcoming of poor speaker recognition performance in noisy environments and combining the advantages of wavelet transform, a method of combining the wavelet transform technology with the traditional characteristic parameter extraction mode is proposed. In this method, the speech signal is decomposed by the wavelet, and then wavelet coefficients are processed by threshold. Only the data above the threshold are retained. The traditional characteristic parameters of little correlation are extracted to use as the input vector of the speaker recognition system. The simulation results indicate that the use of the method can better identify the speaker. A higher recognition rate can be obtained through the wavelet transform first and then the extraction of characteristic parameters. The application of this method greatly improves the performance of the speaker recognition system