Meta releases multilingual speech translation model

Meta has unveiled SeamlessM4T, a groundbreaking speech-to-text model capable of translating nearly 100 languages, aiming to create a universal translator. Termed Massively Multilingual and Multimodal Machine Translation, it can perform speech-to-text and text-to-text translation for 100 input languages, producing 35 output languages. The model is under a Creative Commons CC BY-NC 4.0 license for research use. Alongside, Meta released metadata for its open translation dataset, SeamlessAlign. Unlike previous systems, SeamlessM4T handles complete translation in one step, even identifying code-switching between languages. It tackles gender bias and toxic language, bolstering content moderation and offering potential for broader language support.

‍