Few-step Adversarial Schr\"{o}dinger Bridge for Generative Speech Enhancement

TL;DR

This paper introduces a novel approach combining Schr"odinger Bridge and GANs for speech enhancement, significantly reducing sampling steps while maintaining high-quality denoising and dereverberation performance.

Contribution

It presents a new method that integrates Schr"odinger Bridge with GANs to improve efficiency and performance in generative speech enhancement tasks.

Findings

Outperforms existing models with only one inference step

Maintains high-quality speech enhancement at low SNRs

Reduces sampling steps from over 50 to just 1

Abstract

Deep generative models have recently been employed for speech enhancement to generate perceptually valid clean speech on large-scale datasets. Several diffusion models have been proposed, and more recently, a tractable Schr\"odinger Bridge has been introduced to transport between the clean and noisy speech distributions. However, these models often suffer from an iterative reverse process and require a large number of sampling steps -- more than 50. Our investigation reveals that the performance of baseline models significantly degrades when the number of sampling steps is reduced, particularly under low-SNR conditions. We propose integrating Schr\"odinger Bridge with GANs to effectively mitigate this issue, achieving high-quality outputs on full-band datasets while substantially reducing the required sampling steps. Experimental results demonstrate that our proposed model outperforms existing baselines, even with a single inference step, in both denoising and dereverberation tasks.