A Synergistic Kalman- and Deep Postfiltering Approach to Acoustic Echo Cancellation

TL;DR

This paper presents a novel hybrid approach combining Kalman filtering and deep neural network postfiltering to improve acoustic echo cancellation, especially during abrupt echo path changes, ensuring rapid adaptation and steady-state performance.

Contribution

It introduces a synergistic method that leverages statistical differences and deep learning to enhance adaptive filter robustness in dynamic echo scenarios.

Findings

Rapid reconvergence after echo path changes

Maintains steady-state performance in static scenarios

Effective exploitation of statistical properties for adaptation

Abstract

We introduce a synergistic approach to double-talk robust acoustic echo cancellation combining adaptive Kalman filtering with a deep neural network-based postfilter. The proposed algorithm overcomes the well-known limitations of Kalman filter-based adaptation control in scenarios characterized by abrupt echo path changes. As the key innovation, we suggest to exploit the different statistical properties of the interfering signal components for robustly estimating the adaptation step size. This is achieved by leveraging the postfilter near-end estimate and the estimation error of the Kalman filter. The proposed synergistic scheme allows for rapid reconvergence of the adaptive filter after abrupt echo path changes without compromising the steady state performance achieved by state-of-the-art approaches in static scenarios.