The NiuTrans End-to-End Speech Translation System for IWSLT 2021 Offline Task

Abstract

This paper describes the submission of the NiuTrans end-to-end speech translation system for the IWSLT 2021 offline task, which translates from the English audio to German text directly without intermediate transcription. We use the Transformer-based model architecture and enhance it by Conformer, relative position encoding, and stacked acoustic and textual encoding. To augment the training data, the English transcriptions are translated to German translations. Finally, we employ ensemble decoding to integrate the predictions from several models trained with the different datasets. Combining these techniques, we achieve 33.84 BLEU points on the MuST-C En-De test set, which shows the enormous potential of the end-to-end model.

Related Papers

• Limitations of hybrid systems(2000)
• → A new design method for encoding and decoding circuits(2002)
• Design of fast RS encoding and decoding(2012)
• New kind of 8B/10B encoding and decoding design(2012)
• Study of message encoding and decoding method based on C4ISR system simulation.(2008)