Self-attending RNN for Speech Enhancement to Improve Cross-corpus Generalization

TL;DR

This paper introduces a self-attending RNN architecture for speech enhancement that significantly improves cross-corpus generalization, especially in challenging low SNR conditions, outperforming existing methods.

Contribution

The study proposes a novel self-attending recurrent neural network (ARN) that enhances cross-corpus speech enhancement performance and compares two major approaches, revealing their similar effectiveness.

Findings

ARN outperforms RNNs and dual-path ARNs in low SNR conditions

Complex spectral mapping and time-domain enhancement yield similar results with ARN

A challenging test subset is provided for future benchmarking

Abstract

Deep neural networks (DNNs) represent the mainstream methodology for supervised speech enhancement, primarily due to their capability to model complex functions using hierarchical representations. However, a recent study revealed that DNNs trained on a single corpus fail to generalize to untrained corpora, especially in low signal-to-noise ratio (SNR) conditions. Developing a noise, speaker, and corpus independent speech enhancement algorithm is essential for real-world applications. In this study, we propose a self-attending recurrent neural network, or attentive recurrent network (ARN), for time-domain speech enhancement to improve cross-corpus generalization. ARN comprises of recurrent neural networks (RNNs) augmented with self-attention blocks and feedforward blocks. We evaluate ARN on different corpora with nonstationary noises in low SNR conditions. Experimental results demonstrate that ARN substantially outperforms competitive approaches to time-domain speech enhancement, such as RNNs and dual-path ARNs. Additionally, we report an important finding that the two popular approaches to speech enhancement: complex spectral mapping and time-domain enhancement, obtain similar results for RNN and ARN with large-scale training. We also provide a challenging subset of the test set used in this study for evaluating future algorithms and facilitating direct comparisons.