The HW-TSC's Offline Speech Translation Systems for IWSLT 2021 Evaluation

TL;DR

This paper presents a cascade speech translation system for IWSLT 2021, integrating diarization, ASR, and MT modules, achieving a BLEU score of 24.6 on the test set.

Contribution

It introduces a multi-source trained Transformer-based ASR and a cascade system with diarization for offline speech translation.

Findings

Achieved 24.6 BLEU score on IWSLT 2021 test set.

Utilized multi-source training for ASR with a modified Transformer.

Integrated LIUM SpkDiarization for speaker diarization.

Abstract

This paper describes our work in participation of the IWSLT-2021 offline speech translation task. Our system was built in a cascade form, including a speaker diarization module, an Automatic Speech Recognition (ASR) module and a Machine Translation (MT) module. We directly use the LIUM SpkDiarization tool as the diarization module. The ASR module is trained with three ASR datasets from different sources, by multi-source training, using a modified Transformer encoder. The MT module is pretrained on the large-scale WMT news translation dataset and fine-tuned on the TED corpus. Our method achieves 24.6 BLEU score on the 2021 test set.