A post-filter perceptual convolutive blind source separation approach for speech signal

A frequency domain convolutive blind source separation system concatenated with a post-filter based on masking properties of human auditory system is proposed in this paper to deal with blind speech signal separation. In the post-filter, both time domain and frequency domain masking properties are taken into account. By using the masking filter, the speech spectrum from the convolutive blind source separation system is shaped according to the calculated masking level lo attenuate the residual interference and reverberant effects which remains after imperfect speech separation. Simulation results show that the proposed approach slightly improves the performance of convolutive blind source separation system at no additional cost.