Post-processing for crosstalk cancellation in convolutive BSS outputs based on pitch frequency

Abstract

We propose a novel method to determine crosstalk regions based on the pitch frequency in convolutive blind audio source separation. Since the same pitch frequency in a frame of BSS outputs represents one speaker talking in all separated speech signal, all signals except the speaker's signal contain the crosstalk in that frame. These frames are used in our post-processing algorithm to estimate the crosstalk. Simulation results show that the proposed algorithm successfully suppresses the crosstalk by improving the SIR of BSS outputs about 10.6 dB with adding no distortion to the separated signals.