Unsupervised domain adaptation for speech recognition with unsupervised error correction

TL;DR

This paper introduces an unsupervised error correction method for speech recognition that improves transcription accuracy in unseen domains by using unlabeled data and context-aware models, achieving significant WER reduction.

Contribution

It presents a novel unsupervised error correction approach for domain adaptation in ASR, utilizing pseudo-labeling and context-aware encoder-decoder models, without requiring transcribed target domain data.

Findings

Significant WER reduction over non-adapted systems

Additional 10% relative improvement when combined with other adaptation methods

Effective correction using only unlabeled target domain data

Abstract

The transcription quality of automatic speech recognition (ASR) systems degrades significantly when transcribing audios coming from unseen domains. We propose an unsupervised error correction method for unsupervised ASR domain adaption, aiming to recover transcription errors caused by domain mismatch. Unlike existing correction methods that rely on transcribed audios for training, our approach requires only unlabeled data of the target domains in which a pseudo-labeling technique is applied to generate correction training samples. To reduce over-fitting to the pseudo data, we also propose an encoder-decoder correction model that can take into account additional information such as dialogue context and acoustic features. Experiment results show that our method obtains a significant word error rate (WER) reduction over non-adapted ASR systems. The correction model can also be applied on top of other adaptation approaches to bring an additional improvement of 10% relatively.