Shulou(Shulou.com)11/24 Report--

Thanks to CTOnews.com netizens and soft media users for 1520111 clue delivery! CTOnews.com, August 22 (Xinhua)-- Meta Platforms, the parent company of Facebook, today released an artificial intelligence model, SeamlessM4T, which can translate and transcribe dozens of languages, which allows people who speak different languages to communicate more effectively, helping to develop a human universal language translator.

Meta claims that SeamlessM4T will be open source along with the new translation data set SeamlessAlign, which represents a "major breakthrough" in artificial intelligence-driven "voice-to-voice" and "voice-to-text" areas.

Meta said the model could translate text-to-speech between nearly 100 languages and support full voice-to-speech translation in 35 languages, which previously had to be translated separately in a separate AI model. SeamlessM4T supports:

1. Speech recognition in nearly 100 languages.

2. Speech-to-text translation of nearly 100 input and output languages.

3. Support speech-to-speech translation in nearly 100 input languages and 36 output languages.

4. Support text-to-text translation in nearly 100 languages.

5. Support text-to-speech translation in nearly 100 input languages and 35 output languages.

Currently, Meta will make the model available to the public free of charge in the form of a research license (for non-commercial use only) so that researchers and developers can work on this basis.

According to reports, Meta will also release SeamlessAlign metadata, which is by far the largest open multi-schema translation data set, mining voice and text alignment for a total of 270000 hours.

Mark Zuckerberg said he firmly believes that these tools will facilitate the interaction of users from around the world in the meta-universe, which is his bet on the future of the company.

He also said that opening up the AI ecosystem was good for Meta because the company made more money by effectively crowdsourcing the tasks of creating consumer-oriented tools, rather than charging for access to the model.

It is said that Meta will also integrate these translation and transcriptional AI developments into Facebook, Instagram, WhatsApp, Messenger and Threads.

On the SeamlessM4T model, Meta researchers said in a research paper that they collected 4 million hours of "raw audio" from "a publicly available repository of crawled web data", but CTOnews.com did not find a description of which repository, saying only that the text data came from a dataset created last year, which was extracted from Wikipedia and related websites.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.