Single-channel Speech Dereverberation via Generative Adversarial Training

TL;DR

This paper introduces DeReGAT, a novel single-channel speech dereverberation system utilizing generative adversarial training to enhance speech quality across various reverberant environments.

Contribution

The paper presents a new neural network architecture combined with GAT for improved dereverberation, outperforming existing methods and extending to online processing.

Findings

Outperforms WPE and DNN-based systems in quality.

Effective across wide reverberation ranges.

Online version achieves comparable results to offline.

Abstract

In this paper, we propose a single-channel speech dereverberation system (DeReGAT) based on convolutional, bidirectional long short-term memory and deep feed-forward neural network (CBLDNN) with generative adversarial training (GAT). In order to obtain better speech quality instead of only minimizing a mean square error (MSE), GAT is employed to make the dereverberated speech indistinguishable form the clean samples. Besides, our system can deal with wide range reverberation and be well adapted to variant environments. The experimental results show that the proposed model outperforms weighted prediction error (WPE) and deep neural network-based systems. In addition, DeReGAT is extended to an online speech dereverberation scenario, which reports comparable performance with the offline case.