Task splitting for DNN-based acoustic echo and noise removal

TL;DR

This paper evaluates whether separate or joint modules are more effective for speech enhancement tasks like noise suppression and echo cancellation, finding that separate modules improve performance and reduce distortion.

Contribution

It introduces and compares different implementations of joint versus separate modules for acoustic echo and noise removal, highlighting the benefits of separate modules.

Findings

Separate echo and noise removal modules reduce speech distortion.

Separate modules perform better during double-talk scenarios.

Using separate modules achieves better performance at similar complexity.

Abstract

Neural networks have led to tremendous performance gains for single-task speech enhancement, such as noise suppression and acoustic echo cancellation (AEC). In this work, we evaluate whether it is more useful to use a single joint or separate modules to tackle these problems. We describe different possible implementations and give insights into their performance and efficiency. We show that using a separate echo cancellation module and a module for noise and residual echo removal results in less near-end speech distortion and better performance during double-talk at same complexity.