Personalized speech enhancement combining band-split RNN and speaker attentive module

TL;DR

This paper introduces a personalized speech enhancement model that uses a speaker attentive module to improve extraction of target speech, achieving competitive results in the DNS Challenge 2023.

Contribution

It proposes a novel speaker attentive module for better integration of speaker information into speech enhancement models.

Findings

Achieved a score of 0.529 on track1 and 0.549 on track2 in DNS Challenge 2023.

Demonstrated improved speech enhancement performance with the proposed module.

Abstract

Target speaker information can be utilized in speech enhancement (SE) models to more effectively extract the desired speech. Previous works introduce the speaker embedding into speech enhancement models by means of concatenation or affine transformation. In this paper, we propose a speaker attentive module to calculate the attention scores between the speaker embedding and the intermediate features, which are used to rescale the features. By merging this module in the state-of-the-art SE model, we construct the personalized SE model for ICASSP Signal Processing Grand Challenge: DNS Challenge 5 (2023). Our system achieves a final score of 0.529 on the blind test set of track1 and 0.549 on track2.