The company wrote this in a blog post The SeamlessM4T model can support text-to-speech translation in nearly 100 languages, as well as full speech-to-speech translation for 35 languages, Combine technologies that were previously available only in separate models.
CEO Mark Zuckerberg said he envisions such tools to facilitate interactions between users from all over the world in the metaverse. The blog post reveals that Meta makes the form publicly available for non-commercial use.
The world’s largest social media company has released several mostly free AI models this year, including a large-scale language model called Llama that poses a serious challenge to Microsoft-backed OpenAI and proprietary models marketed by Google.
According to Zuckerberg, an open AI ecosystem will benefit Meta, The company is poised to achieve more by crowdsourcing consumer tools for its social platforms.
For the SeamlessM4T model, Meta researchers said in a research paper that they collected 4 million hours of “audio training data from raw audio from crawled publicly available web data,” without specifying the repository. A Meta spokesperson did not respond to questions about the origin of the audio data.
Cover image source: Getty Images
