ComSL: A Composite Speech-Language Model for End-to-End Speech-to-Text Translation

TL;DR

ComSL is a novel composite speech-language model that efficiently combines pretrained speech and language models through multi-task learning, achieving state-of-the-art results in multilingual speech-to-text translation.

Contribution

It introduces a composite architecture with cross-modality transfer learning for end-to-end speech translation, reducing data and computational demands.

Findings

Achieved a new state-of-the-art BLEU score of 31.5 on CoVoST2.

Effectively integrates speech and language models for multilingual translation.

Demonstrated data-efficient training with improved performance.

Abstract

Joint speech-language training is challenging due to the large demand for training data and GPU consumption, as well as the modality gap between speech and language. We present ComSL, a speech-language model built atop a composite architecture of public pretrained speech-only and language-only models and optimized data-efficiently for spoken language tasks. Particularly, we propose to incorporate cross-modality learning into transfer learning and conduct them simultaneously for downstream tasks in a multi-task learning manner. Our approach has demonstrated effectiveness in end-to-end speech-to-text translation tasks, achieving a new state-of-the-art average BLEU score of 31.5 on the multilingual speech to English text translation task for 21 languages, as measured on the public CoVoST2 evaluation set.