Is Phase Really Needed for Weakly-Supervised Dereverberation ?

TL;DR

This paper investigates the importance of reverberant phase information in weakly-supervised speech dereverberation, showing that phase is largely uninformative and can be excluded to improve model performance.

Contribution

The study demonstrates that reverberant phase provides limited useful information for dereverberation, challenging the assumption that phase is essential in weakly-supervised approaches.

Findings

Excluding wet phase improves dereverberation performance.

Reverberant phase acts as noise, not useful information.

Late reverberation perturbs phase with white noise.

Abstract

In unsupervised or weakly-supervised approaches for speech dereverberation, the target clean (dry) signals are considered to be unknown during training. In that context, evaluating to what extent information can be retrieved from the sole knowledge of reverberant (wet) speech becomes critical. This work investigates the role of the reverberant (wet) phase in the time-frequency domain. Based on Statistical Wave Field Theory, we show that late reverberation perturbs phase components with white, uniformly distributed noise, except at low frequencies. Consequently, the wet phase carries limited useful information and is not essential for weakly supervised dereverberation. To validate this finding, we train dereverberation models under a recent weak supervision framework and demonstrate that performance can be significantly improved by excluding the reverberant phase from the loss function.