Self-Training for End-to-End Speech Recognition

TL;DR

This paper demonstrates that self-training with pseudo-labels, combined with filtering and ensemble techniques, significantly improves end-to-end speech recognition accuracy, especially in noisy conditions, by leveraging unlabeled data.

Contribution

It introduces a novel self-training approach with filtering and ensemble methods that substantially enhances speech recognition performance over previous methods.

Findings

33.9% relative WER reduction on LibriSpeech with ensemble and filtering

Recovers 59.3% of the gap to an oracle model in noisy speech

Outperforms previous approaches in semi-supervised speech recognition

Abstract

We revisit self-training in the context of end-to-end speech recognition. We demonstrate that training with pseudo-labels can substantially improve the accuracy of a baseline model. Key to our approach are a strong baseline acoustic and language model used to generate the pseudo-labels, filtering mechanisms tailored to common errors from sequence-to-sequence models, and a novel ensemble approach to increase pseudo-label diversity. Experiments on the LibriSpeech corpus show that with an ensemble of four models and label filtering, self-training yields a 33.9% relative improvement in WER compared with a baseline trained on 100 hours of labelled data in the noisy speech setting. In the clean speech setting, self-training recovers 59.3% of the gap between the baseline and an oracle model, which is at least 93.8% relatively higher than what previous approaches can achieve.