Gradient Remedy for Multi-Task Learning in End-to-End Noise-Robust Speech Recognition

TL;DR

This paper introduces a gradient remedy method to address gradient interference in multi-task learning for noise-robust speech recognition, leading to significant WER reductions on benchmark datasets.

Contribution

It proposes a novel gradient remedy approach that projects and rescales task gradients to improve multi-task learning in noise-robust speech recognition.

Findings

Achieves 9.3% and 11.1% relative WER reduction on RATS and CHiME-4 datasets.

Effectively resolves gradient interference between speech enhancement and recognition tasks.

Enhances multi-task learning performance in noisy speech recognition scenarios.

Abstract

Speech enhancement (SE) is proved effective in reducing noise from noisy speech signals for downstream automatic speech recognition (ASR), where multi-task learning strategy is employed to jointly optimize these two tasks. However, the enhanced speech learned by SE objective may not always yield good ASR results. From the optimization view, there sometimes exists interference between the gradients of SE and ASR tasks, which could hinder the multi-task learning and finally lead to sub-optimal ASR performance. In this paper, we propose a simple yet effective approach called gradient remedy (GR) to solve interference between task gradients in noise-robust speech recognition, from perspectives of both angle and magnitude. Specifically, we first project the SE task's gradient onto a dynamic surface that is at acute angle to ASR gradient, in order to remove the conflict between them and assist in ASR optimization. Furthermore, we adaptively rescale the magnitude of two gradients to prevent the dominant ASR task from being misled by SE gradient. Experimental results show that the proposed approach well resolves the gradient interference and achieves relative word error rate (WER) reductions of 9.3% and 11.1% over multi-task learning baseline, on RATS and CHiME-4 datasets, respectively. Our code is available at GitHub.