New frequency domain post-filters for noise cancellation using two closely spaced microphones

This paper addresses the problem of speech enhancement in a moving car through a blind source separation (BSS) scheme involving two closely spaced microphones. We propose two frequency domain methods to reduce the distortion caused by the forward BSS structure, which is most important when microphones are closely spaced. Both methods aim at estimating post-filters to compensate for the distortion by equalization. The first method is based on an open-loop estimation. The second one uses a closed-loop adaptive algorithm to make the estimation recursively. Simulation results show that the second method is more efficient than the first one, and both of them allow to correct the signal distortion caused by the BSS forward structure.