Model-based Single-Channel Dereverberation in Noisy Acoustical Environments

This paper illustrates a new system for recovering clean speech signals from noisy acoustical environments using one microphone. At the beginning of this paper, we propose an assumption that the background noise is comprised of reverberant noise and direct-path noise. And a novel late reverberant spectral variance (LRSV) estimator is generated referring to this assumption, which can be used in the noisy acoustical environments. What’s more, the shape parameters of this LRSV estimator are updated by taking some frames of previous late reverberation into account. At last, a new spectral process system is developed to help making the LRSV estimator more efficient. The experimental results show the benefits of our new system when it is used to reduce the interference in both noise-free and noisy acoustical environments.