UniSpeech: Unified Speech Representation Learning with Labeled and Unlabeled Data
This work addresses the challenge of improving speech recognition performance in cross-lingual and domain-shift scenarios, representing an incremental advance by integrating existing methods.
The paper tackles the problem of learning speech representations that generalize across languages and domains by proposing UniSpeech, a unified pre-training approach combining supervised and self-supervised learning. The results show maximum relative reductions of 13.4% in phone error rate for cross-lingual tasks and 6% in word error rate for domain-shift tasks.
In this paper, we propose a unified pre-training approach called UniSpeech to learn speech representations with both unlabeled and labeled data, in which supervised phonetic CTC learning and phonetically-aware contrastive self-supervised learning are conducted in a multi-task learning manner. The resultant representations can capture information more correlated with phonetic structures and improve the generalization across languages and domains. We evaluate the effectiveness of UniSpeech for cross-lingual representation learning on public CommonVoice corpus. The results show that UniSpeech outperforms self-supervised pretraining and supervised transfer learning for speech recognition by a maximum of 13.4% and 17.8% relative phone error rate reductions respectively (averaged over all testing languages). The transferability of UniSpeech is also demonstrated on a domain-shift speech recognition task, i.e., a relative word error rate reduction of 6% against the previous approach.