MADRID, 22 Aug. (Portaltic/EP) –
Goal has presented this Tuesday a new artificial intelligence (AI) model that performs translations and transcriptions of almost a hundred languages as part of their work for the development of a universal translator.
SeamlessM4T follow a single system approach to reduce errors and delays in the translation process, and takes advantage of advances made in other company models, such as No Language Left Behind, the universal speech translator, and massive multilingual voice models.
Specifically, this multimodal and multilingual model can recognize speech in nearly 100 languages, translate speech to text in nearly 100 input and output languages, translate speech to speech in nearly 100 input and 36 output languages, translate text to text in nearly 100 languages and translate text to speech in one hundred input and 35 output languages.
“Build a universal language translator, like the fictional Fish of Babel from [la novela] Hitchhiker’s Guide to the Galaxy, it’s a challenge because existing speech-to-speech and speech-to-text systems only cover a small fraction of the world’s languages. But we believe that the work we are announcing today is an important step forward on this journey,” the company says on its AI blog.
Alongside this model, Meta also provided the open multimodal translation dataset SeamlessAlignwith a total of 270,000 hours of voice and text alignments.