T-Modules: Translation Modules for Zero-Shot Cross-Modal Machine Translation
This enables zero-shot translation across languages and modalities without labeled data, addressing a key bottleneck in multimodal AI for applications like speech-to-speech translation.
The paper tackles zero-shot cross-modal translation between speech and text by encoding multilingual data into a joint fixed-size representation space and decoding it without cross-modal labeled data, achieving competitive results and significantly improving state-of-the-art on Must-C for zero-shot speech translation.
We present a new approach to perform zero-shot cross-modal transfer between speech and text for translation tasks. Multilingual speech and text are encoded in a joint fixed-size representation space. Then, we compare different approaches to decode these multimodal and multilingual fixed-size representations, enabling zero-shot translation between languages and modalities. All our models are trained without the need of cross-modal labeled translation data. Despite a fixed-size representation, we achieve very competitive results on several text and speech translation tasks. In particular, we significantly improve the state-of-the-art for zero-shot speech translation on Must-C. Incorporating a speech decoder in our framework, we introduce the first results for zero-shot direct speech-to-speech and text-to-speech translation.