Meta launches AI model for speech and text translations

Meta has launched an AI model that will enable speech and text translations for up to 100 languages.

Meta announced the launch of SeamlessM4T on Tuesday, a multilingual AI model that will translate and transcribe across speech and text. Platform users will be able to perform speech-to-text, speech-to-speech, text-to-speech, and text-to-text translations.

SeamlessM4T is released under a research license, allowing researchers and developers to contribute to further development of the technology. The company is also sharing the metadata of SeamlessAlign, the biggest open multimodal translation dataset, totaling 270,000 hours of mined speech and text alignments.

Building technology for AI-based translations

The new release builds upon Meta’s previous efforts to master AI-assisted translations. Last year, the company released No Language Left Behind (NLLB), a text-to-text machine translation model that supports 200 languages.

The company also announced the creation of its Universal Speech Translator, the first AI speech-to-speech translation system for the Hokkien language. This orally-based language is extensively spoken among the Chinese diaspora and is characterized by the absence of a written form.

Earlier this year, Meta released a Massively Multilingual Speech AI research model, which can identify more than 4000 spoken languages.