Researchers have developed an AI-powered translation model capable of real-time speech-to-speech translation across 101 languages. 

Dubbed a modern analogue of the fictional Babel Fish from The Hitchhiker's Guide to the Galaxy, this model facilitates instant, multilingual communication without requiring intermediary text conversion.  

Unlike traditional translation systems, which rely on a multi-step process involving transcription and subsequent translation, Meta’s SEAMLESSM4T processes speech directly into another language. This eliminates delays and increases accuracy. 

A study published in Nature highlights that SEAMLESSM4T achieves up to 23 per cent higher translation accuracy than existing models. 

Additionally, it filters background noise and adapts to speaker variations, enhancing its utility in diverse real-world settings.  

“By bridging gaps in language coverage and reducing the reliance on English as an intermediary, this model pushes the boundaries of speech translation,” says Dr Claudia Singh, a member of the Seamless Communication Team at Meta.

SEAMLESSM4T supports direct translation of speech into 36 languages and expands into text translation and recognition for up to 96 languages.  

The resource is freely available for public, non-commercial use. Developers and researchers are encouraged to utilise the tool to advance inclusive multilingual communication technologies.  

This email address is being protected from spambots. You need JavaScript enabled to view it. CareerSpot News