Can be tried online
Meta* introduced the SeamlessM4T artificial intelligence model, which can transcribe speech in almost 100 languages and translate it.
The company itself says it’s a breakthrough in AI-powered speech-to-speech and speech-to-text.
Our unified model provides on-demand translations that enable people of different languages to communicate more effectively. SeamlessM4T implicitly recognizes source languages without the need for a separate language identification model
Judging by the last sentence, we are talking about an AI model that was initially trained both to recognize the language and convert it to text, and to translate. That is, these are not two separate models in one software.
Artificial intelligence that recognizes speech and translates it into 100 other languages.
During the development process, Meta* used about “tens of billions” of sentences and 4 million hours of speech. In addition, the researchers matched 443,000 hours of speech to texts and created 29,000 hours of speech-to-speech matches that taught SeamlessM4T to transcribe speech to text, translate text, generate speech from text, and even translate speech.
At the same time, Meta understands that its model is not ideal, therefore it does not recommend using SeamlessM4T to translate long forms and any documents, such as those recognized by government agencies and translation authorities. Meta also discourages the use of SeamlessM4T for medical or legal purposes.
You can test the work of AI at this link .
* The Meta company is recognized as extremist in Russia and banned.
