Atss-Net: Target Speaker Separation via Attention-based Neural Network

TL;DR

This paper introduces Atss-Net, an attention-based neural network for target speaker separation that outperforms existing models like VoiceFilter while using fewer parameters, and also shows promise in speech enhancement.

Contribution

The paper presents a novel attention-based neural network architecture for speaker separation that is more efficient and effective than previous CNN-LSTM models.

Findings

Atss-Net outperforms VoiceFilter in speaker separation tasks.

Atss-Net uses fewer parameters than comparable models.

The model shows promising results in speech enhancement.

Abstract

Recently, Convolutional Neural Network (CNN) and Long short-term memory (LSTM) based models have been introduced to deep learning-based target speaker separation. In this paper, we propose an Attention-based neural network (Atss-Net) in the spectrogram domain for the task. It allows the network to compute the correlation between each feature parallelly, and using shallower layers to extract more features, compared with the CNN-LSTM architecture. Experimental results show that our Atss-Net yields better performance than the VoiceFilter, although it only contains half of the parameters. Furthermore, our proposed model also demonstrates promising performance in speech enhancement.