Universal Speech Enhancement with Regression and Generative Mamba

TL;DR

The paper introduces USEMamba, a versatile speech enhancement model capable of handling various distortions and languages, achieving high generalization with a mix of regression and generative techniques.

Contribution

It presents a novel state-space model for universal speech enhancement that unifies multiple tasks and conditions, including a generative variant for missing content inference.

Findings

Achieved 2nd place in Interspeech 2025 URGENT Challenge Track 1.

Demonstrated strong generalization across diverse distortions and languages.

Effective combination of regression and generative modeling for different enhancement scenarios.

Abstract

The Interspeech 2025 URGENT Challenge aimed to advance universal, robust, and generalizable speech enhancement by unifying speech enhancement tasks across a wide variety of conditions, including seven different distortion types and five languages. We present Universal Speech Enhancement Mamba (USEMamba), a state-space speech enhancement model designed to handle long-range sequence modeling, time-frequency structured processing, and sampling frequency-independent feature extraction. Our approach primarily relies on regression-based modeling, which performs well across most distortions. However, for packet loss and bandwidth extension, where missing content must be inferred, a generative variant of the proposed USEMamba proves more effective. Despite being trained on only a subset of the full training data, USEMamba achieved 2nd place in Track 1 during the blind test phase, demonstrating strong generalization across diverse conditions.