Improving Streaming Transformer Based ASR Under a Framework of Self-supervised Learning

TL;DR

This paper enhances streaming transformer-based automatic speech recognition by introducing a two-stage fine-tuning method combining knowledge distillation and self-training within a self-supervised learning framework, achieving significant WER reductions.

Contribution

It proposes a novel two-stage training approach for streaming transformers that integrates knowledge distillation and self-training under self-supervised learning, improving ASR performance.

Findings

Achieves 16.3% relative WER reduction on Librispeech noisy test set.

Attains WERs of 3.5/8.7 on Librispeech clean/noisy test sets using limited labeled data.

Demonstrates effectiveness of the proposed method in streaming ASR scenarios.

Abstract

Recently self-supervised learning has emerged as an effective approach to improve the performance of automatic speech recognition (ASR). Under such a framework, the neural network is usually pre-trained with massive unlabeled data and then fine-tuned with limited labeled data. However, the non-streaming architecture like bidirectional transformer is usually adopted by the neural network to achieve competitive results, which can not be used in streaming scenarios. In this paper, we mainly focus on improving the performance of streaming transformer under the self-supervised learning framework. Specifically, we propose a novel two-stage training method during fine-tuning, which combines knowledge distilling and self-training. The proposed training method achieves 16.3% relative word error rate (WER) reduction on Librispeech noisy test set. Finally, by only using the 100h clean subset of Librispeech as the labeled data and the rest (860h) as the unlabeled data, our streaming transformer based model obtains competitive WERs 3.5/8.7 on Librispeech clean/noisy test sets.