Neural Network-augmented Kalman Filtering for Robust Online Speech Dereverberation in Noisy Reverberant Environments

TL;DR

This paper introduces a neural network-augmented Kalman filtering approach for online speech dereverberation that enhances robustness and performance in noisy, reverberant environments by correcting filter variation estimates in a data-driven manner.

Contribution

It proposes a novel neural network-augmented Kalman filtering method for online speech dereverberation that improves robustness and reduces distortions in noisy conditions.

Findings

Outperforms DNN-supported recursive least squares WPE in noisy scenarios

Provides strong dereverberation and denoising results

Increases robustness to noisy reverberant environments

Abstract

In this paper, a neural network-augmented algorithm for noise-robust online dereverberation with a Kalman filtering variant of the weighted prediction error (WPE) method is proposed. The filter stochastic variations are predicted by a deep neural network (DNN) trained end-to-end using the filter residual error and signal characteristics. The presented framework allows for robust dereverberation on a single-channel noisy reverberant dataset similar to WHAMR!. The Kalman filtering WPE introduces distortions in the enhanced signal when predicting the filter variations from the residual error only, if the target speech power spectral density is not perfectly known and the observation is noisy. The proposed approach avoids these distortions by correcting the filter variations estimation in a data-driven way, increasing the robustness of the method to noisy scenarios. Furthermore, it yields a strong dereverberation and denoising performance compared to a DNN-supported recursive least squares variant of WPE, especially for highly noisy inputs.