Posterior sampling algorithms for unsupervised speech enhancement with recurrent variational autoencoder

TL;DR

This paper introduces efficient sampling algorithms for unsupervised speech enhancement using RVAE, improving computational efficiency and robustness over traditional variational inference methods and supervised approaches.

Contribution

It proposes Langevin dynamics and Metropolis-Hasting sampling techniques to replace variational inference in RVAE-based speech enhancement, reducing complexity and enhancing performance.

Findings

Sampling methods outperform VEM in efficiency and accuracy

Proposed algorithms generalize well to mismatched conditions

Outperform supervised diffusion model approaches

Abstract

In this paper, we address the unsupervised speech enhancement problem based on recurrent variational autoencoder (RVAE). This approach offers promising generalization performance over the supervised counterpart. Nevertheless, the involved iterative variational expectation-maximization (VEM) process at test time, which relies on a variational inference method, results in high computational complexity. To tackle this issue, we present efficient sampling techniques based on Langevin dynamics and Metropolis-Hasting algorithms, adapted to the EM-based speech enhancement with RVAE. By directly sampling from the intractable posterior distribution within the EM process, we circumvent the intricacies of variational inference. We conduct a series of experiments, comparing the proposed methods with VEM and a state-of-the-art supervised speech enhancement approach based on diffusion models. The results reveal that our sampling-based algorithms significantly outperform VEM, not only in terms of computational efficiency but also in overall performance. Furthermore, when compared to the supervised baseline, our methods showcase robust generalization performance in mismatched test conditions.