Meta Introduces SeamlessM4T, an AI Model for Translations

Meta is planning to introduce SeamlessM4T, an all-in-one multimodal and multilingual artificial intelligence model for speech and text translations.

SeamlessM4T builds on Meta's existing Universal Speech Translator, SpeechMatrix, and Massively Multilingual Speech to create a single model that can handle multilingual and multimodal translations. It performs automatic speech recognition and speech-to-text, speech-to-speech, text-to-text, and text-to-speech translations in nearly 100 languages via a single AI model.

The single system approach reduces errors and delays, according to Meta, which plans to release SeamlessM4T along with the metadata of SeamlessAlign, its open multimodal translation dataset containing more than 270,000 hours of mined speech and text alignment, in the coming weeks.