Speech Enhancement based on Residual Noise Shaping

In this paper, we propose a novel approach to speech enhancement, which incorporates a new criterion based on residual noise shaping. In the proposed approach, our goal is to make the residual noise perceptually comfortable although the power of the residual noise is relatively high. In contrast to the conventional techniques, the proposed approach regulates not only the power of the signal distortion and residual noise but also the spectral shape of the residual noise. A predetermined 'comfort noise' is provided as a target for the spectral shaping. Three different versions of enhancement algorithms adopting the proposed criterion are presented. Subjective listening test results show that the proposed algorithm outperforms the conventional spectral enhancement techniques which are based on soft decision and the noise suppression module implemented in IS-893 Selectable Mode Vocoder.