eduzhai > Applied Sciences > Engineering >

Self-Training for End-to-End Speech Translation

  • Save

... pages left unread,continue reading

Document pages: 5 pages

Abstract: One of the main challenges for end-to-end speech translation is datascarcity. We leverage pseudo-labels generated from unlabeled audio by a cascadeand an end-to-end speech translation model. This provides 8.3 and 5.7 BLEUgains over a strong semi-supervised baseline on the MuST-C English-French andEnglish-German datasets, reaching state-of-the art performance. The effect ofthe quality of the pseudo-labels is investigated. Our approach is shown to bemore effective than simply pre-training the encoder on the speech recognitiontask. Finally, we demonstrate the effectiveness of self-training by directlygenerating pseudo-labels with an end-to-end model instead of a cascade model.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...