Unsupervised Cross-lingual Representation Learning for Speech Recognition

TL;DR

This paper introduces XLSR, a cross-lingual speech representation model trained on multiple languages' raw speech data, significantly improving speech recognition accuracy across various benchmarks.

Contribution

We propose XLSR, a novel multilingual pretraining approach based on wav2vec 2.0, enabling a single model to perform well across many languages, especially benefiting low-resource settings.

Findings

72% phoneme error rate reduction on CommonVoice

16% word error rate improvement on BABEL

Shared latent speech representations across languages

Abstract

This paper presents XLSR which learns cross-lingual speech representations by pretraining a single model from the raw waveform of speech in multiple languages. We build on wav2vec 2.0 which is trained by solving a contrastive task over masked latent speech representations and jointly learns a quantization of the latents shared across languages. The resulting model is fine-tuned on labeled data and experiments show that cross-lingual pretraining significantly outperforms monolingual pretraining. On the CommonVoice benchmark, XLSR shows a relative phoneme error rate reduction of 72% compared to the best known results. On BABEL, our approach improves word error rate by 16% relative compared to a comparable system. Our approach enables a single multilingual speech recognition model which is competitive to strong individual models. Analysis shows that the latent discrete speech representations are shared across languages with increased sharing for related languages. We hope to catalyze research in low-resource speech understanding by releasing XLSR-53, a large model pretrained in 53 languages.