On the application of minimum noise tracking to cancel cosine shaped residual noise

It has been shown recently that for coherence based dual microphone array speech enhancement systems, cross-spectral subtraction is an efficient technique aimed to reduce the correlated noise components. The zero-phase filtering criterion employed in these methods is derived from the standard coherence function that is modified to incorporate the noise cross power spectrum between the two channels. However, there has been limited success at applying coherence based filters when speech processing is carried out under relatively harsh acoustic conditions (SNR below -5dB) or when the speech and noise sources are closely spaced. We propose an alternative method that is effective, and that attempts to use a phase-based filtering criterion by substituting the cross power spectrum of the noisy signals received on the two channels by its real part. Then, a variant of the running minimum noise tracking procedure is applied on the estimated speech spectrum as an adaptive postfiltering to reduce the cosine shaped power spectrum of the remaining residual musical noise to a minimum spectral floor. Using that adaptive postfilter, a softdecision scheme is implemented to control the amount of noise suppression. Our preliminary results based on experiments conducted on real speech signals show an improved performance of the proposed method over the coherence based approaches. These results also show that it performs well on speech while producing less spectral distortion even in severe noisy conditions.