Single-Microphone Speech Enhancement by Skewness Maximization and Spectral Subtraction

Abstract:
A signal received by a microphone is corrupted by additive noise and distorted by room reverberation. Reverberation consists of two main components, early and late reverberation, both of which degrade speech quality. Thus it is desirable to reduce the effects of noise and reverberation. In this paper, we propose a single microphone solution to deal with these problems. Inverse filtering is used to reduce the early reverberation effects by maximizing the skewness of the LP residual. Then, spectral subtraction is used to reduce both the noise and remaining late reverberation effects. This approach introduces negligible musical noise compared to the method of Wu and Wang. Performance results are given which demonstrate that our approach provides superior performance in a variety of noisy reverberant conditions. Index Terms — Single-microphone, early and late reverberation, skewness, inverse filtering, spectral subtraction