Robust wav2vec 2.0: Analyzing Domain Shift in Self-Supervised Pre-Training

TL;DR

This paper investigates how domain differences affect self-supervised speech model performance and demonstrates that using target domain data during pre-training significantly enhances robustness and generalization across various domains.

Contribution

It introduces an analysis of domain shift in self-supervised speech learning and shows that in-domain pre-training reduces performance gaps and improves cross-domain generalization.

Findings

Pre-training on target domain data greatly improves performance.

Using multiple domains during pre-training enhances unseen domain generalization.

In-domain pre-training reduces the performance gap by up to 73%.

Abstract

Self-supervised learning of speech representations has been a very active research area but most work is focused on a single domain such as read audio books for which there exist large quantities of labeled and unlabeled data. In this paper, we explore more general setups where the domain of the unlabeled data for pre-training data differs from the domain of the labeled data for fine-tuning, which in turn may differ from the test data domain. Our experiments show that using target domain data during pre-training leads to large performance improvements across a variety of setups. On a large-scale competitive setup, we show that pre-training on unlabeled in-domain data reduces the gap between models trained on in-domain and out-of-domain labeled data by 66%-73%. This has obvious practical implications since it is much easier to obtain unlabeled target domain data than labeled data. Moreover, we find that pre-training on multiple domains improves generalization performance on domains not seen during training. Code and models will be made available at https://github.com/pytorch/fairseq.