U-DREAM: Unsupervised Dereverberation guided by a Reverberation Model

TL;DR

This paper introduces U-DREAM, an unsupervised dereverberation method that trains deep neural networks using only reverberant signals and a reverberation model, eliminating the need for paired dry data.

Contribution

It proposes a sequential learning strategy guided by a reverberation matching loss, enabling effective dereverberation with minimal labeled data.

Findings

Achieves superior dereverberation performance with only 100 labeled samples.

Outperforms baseline methods in low-resource scenarios.

Demonstrates practicality of unsupervised training for dereverberation.

Abstract

This paper explores the outcome of training state-of-the-art dereverberation models with supervision settings ranging from weakly-supervised to virtually unsupervised, relying solely on reverberant signals and an acoustic model for training. Most of the existing deep learning approaches typically require paired dry and reverberant data, which are difficult to obtain in practice. We develop instead a sequential learning strategy motivated by a maximum-likelihood formulation of the dereverberation problem, wherein acoustic parameters and dry signals are estimated from reverberant inputs using deep neural networks, guided by a reverberation matching loss. Our most data-efficient variant requires only 100 reverberation-parameter-labeled samples to outperform an unsupervised baseline, demonstrating the effectiveness and practicality of the proposed method in low-resource scenarios.