Meta SeamlessM4T

What is Meta SeamlessM4T?

SeamlessM4T is a foundational model developed by Meta that handles both speech and text translation across many languages and tasks. Its architecture, called multitask UnitY, is designed to tackle a range of translation jobs, including recognizing speech, translating text, and synthesizing speech.

Who created Meta SeamlessM4T?

SeamlessM4T was developed by a team of researchers and engineers at Meta. Key contributors include Bapi Akula, Pierre Andrews, and Loïc Barrault, among many others. The project was a collaborative effort to build a single foundational model capable of handling speech and text translation across many languages.

Open science and licensing

Meta has made the model publicly available under the CC BY-NC 4.0 license. Alongside the model, the team released a large dataset called SeamlessAlign, supporting further research and development in multilingual translation.

Summary

Meta SeamlessM4T is a multitask speech and text translation model built on the UnitY architecture, released openly under the CC BY-NC 4.0 license together with the SeamlessAlign dataset. It reflects Meta’s commitment to open science in multilingual AI research.