Scheduled Interleaved Speech-Text Training for Speech-to-Speech Translation with LLMs

TL;DR

This paper introduces a scheduled interleaved speech-text training method for speech-to-speech translation using LLMs, which gradually shifts from text to speech modality to improve translation quality, especially for low-resource languages.

Contribution

The study proposes a novel training strategy that interleaves speech and text units with a gradual transition, enhancing modality adaptation in speech-to-speech translation models.

Findings

Improved translation performance across multiple languages.

Enhanced adaptation from text to speech modality.

Effective for low-resource language translation.

Abstract

Speech-to-speech translation (S2ST) has been advanced with large language models (LLMs), which are fine-tuned on discrete speech units. In such approaches, modality adaptation from text to speech has been an issue. LLMs are trained on text-only data, which presents challenges to adapt them to speech modality with limited speech-to-speech data. To address the training difficulty, we propose scheduled interleaved speech--text training in this study. We use interleaved speech--text units instead of speech units during training, where aligned text tokens are interleaved at the word level. We gradually decrease the ratio of text as training progresses, to facilitate progressive modality adaptation from text to speech. We conduct experimental evaluations by fine-tuning LLaMA3.2-1B for S2ST on the CVSS dataset. We show that the proposed method consistently improves the translation performances, especially for languages with limited training data.