Effect of the perceptual weighting by spectral shaping of residual noise on time-domain multichannel noise reduction

Abstract

This letter studies the effect of the perceptual weighting by spectral shaping of residual noise, a technique that has been successfully used in signal-channel noise reduction, on multichannel noise reduction in the time domain. It is known that multichannel methods can achieve noise reduction without introducing any speech distortion compared with single-channel methods. This letter shows, however, that the perceptual weighting by spectral shaping of residual noise will have no effect on multichannel noise reduction if no speech distortion is expected. It is useful only if some speech distortion is tolerated. Moreover, the effect is more significant with the increase of speech distortion.