The Meta company, as part of its work on developing a universal translator, introduced a new model of artificial intelligence (AI) that performs translation and transcription in almost a hundred languages.
This model, called “SeamlessM4T”, follows a uniform system approach to reduce errors and delays in translation processes.
In addition, it leverages the advances of other business models, such as B. No Language Left Behind, the universal language translator, and massive multilingual language models.
The new multimodal and multilingual model can recognize speech in almost a hundred languages and translate from speech to text in nearly a hundred input and output languages, the ADP agency reported.
Also within the scope of your possibilities, you can translate from language to language in almost a hundred input languages and 36 output languages, translate text to text in almost a hundred languages, and translate text to speech in a hundred input and 35 output languages.
“Building in Universal language translators, like the fictional fish of Babel from (the novel) The Hitchhiker’s Guide to the Galaxy, is challenging because existing speech-to-speech and speech-to-text systems only cover a small fraction of the world’s languages,” Meta said on his AI blog.
“But we believe the work we’re announcing today is an important step forward in that journey, they added.
In addition to this model, Meta also provided the open multimodal translation dataset SeamlessAlign, containing a total of 270,000 hours of language and text alignments.