MUSE: Flexible Voiceprint Receptive Fields and Multi-Path Fusion Enhanced Taylor Transformer for U-Net-based Speech Enhancement

TL;DR

This work tackles the challenge of achieving strong speech enhancement with a lightweight model. It introduces MUSE, a U-Net-based architecture that integrates Deformable Embedding and a Multi-path Enhanced Taylor Transformer (MET) to learn flexible voiceprint receptive fields and effective feature fusion while maintaining a tiny footprint of $0.51M$ parameters. The MET block combines Taylor multi-head attention with channel and spatial attention branches, fused via a Simple Gate, and is complemented by a Dense Convolution Codec to broaden receptive fields. On VoiceBank+DEMAND, MUSE delivers competitive objective metrics with significantly reduced training and deployment costs, demonstrating practical potential for real-time, low-resource speech enhancement applications.

Abstract

Achieving a balance between lightweight design and high performance remains a challenging task for speech enhancement. In this paper, we introduce Multi-path Enhanced Taylor (MET) Transformer based U-net for Speech Enhancement (MUSE), a lightweight speech enhancement network built upon the Unet architecture. Our approach incorporates a novel Multi-path Enhanced Taylor (MET) Transformer block, which integrates Deformable Embedding (DE) to enable flexible receptive fields for voiceprints. The MET Transformer is uniquely designed to fuse Channel and Spatial Attention (CSA) branches, facilitating channel information exchange and addressing spatial attention deficits within the Taylor-Transformer framework. Through extensive experiments conducted on the VoiceBank+DEMAND dataset, we demonstrate that MUSE achieves competitive performance while significantly reducing both training and deployment costs, boasting a mere 0.51M parameters.

Figures (2)

• Figure 1: The overall architecture of MUSE.
• Figure 2: MET Transformer.