Speech Enhancement with Fullband-Subband Cross-Attention Network

TL;DR

This paper introduces FS-CANet, a novel speech enhancement model that employs fullband-subband cross-attention to better fuse global and local information, outperforming existing methods on standard datasets.

Contribution

The paper proposes a fullband-subband cross-attention module and integrates it into FullSubNet, enhancing interaction between fullband and subband features and reducing model size with TCN blocks.

Findings

FS-CANet outperforms state-of-the-art speech enhancement models.

The cross-attention mechanism effectively fuses global and local information.

Model size is reduced while maintaining high performance.

Abstract

FullSubNet has shown its promising performance on speech enhancement by utilizing both fullband and subband information. However, the relationship between fullband and subband in FullSubNet is achieved by simply concatenating the output of fullband model and subband units. It only supplements the subband units with a small quantity of global information and has not considered the interaction between fullband and subband. This paper proposes a fullband-subband cross-attention (FSCA) module to interactively fuse the global and local information and applies it to FullSubNet. This new framework is called as FS-CANet. Moreover, different from FullSubNet, the proposed FS-CANet optimize the fullband extractor by temporal convolutional network (TCN) blocks to further reduce the model size. Experimental results on DNS Challenge - Interspeech 2021 dataset show that the proposed FS-CANet outperforms other state-of-the-art speech enhancement approaches, and demonstrate the effectiveness of fullband-subband cross-attention.