Abstract: Critical band wavelet transform that relates to auditory feature is used to analyze overlapping speech signals. Based on critical band wavelet transform, a method is presented to solve the problem of overlapping speech signals' pitch detection and separation, then a set of harmonic of each source speech can be obtained by using the frequency selection characteristic of digital multivibrator. So we can separate the multisource speech signals from overlapping speech. The experiment results indicate that the overlapping speech separation method is very efficient.