#artificial intelligence
Nov 2025
OmniFusion: Simultaneous Multilingual Multimodal Translations via Modular Fusion
An end-to-end approach that fuses MMFMs with translation LLMs, enabling joint end-to-end training and achieving a 1-second latency reduction in SimulST compared to cascaded pipelines and also improves the overall translation quality is proposed.
Sai Koneru, Matthias Huck, Jan Niehues
· arXiv.org · 0 citations