Thursday, June 20, 2024 Banner
HomeSoftwareAIMeta Releases SeamlessM4T AI Model to Translate Languages in Real-Time Go.

Meta Releases SeamlessM4T AI Model to Translate Languages in Real-Time Go.

Facebook’s parent company, Meta Platforms, unveiled an AI model on Tuesday that can interpret and record speech in a variety of languages. This advancement could be used as a building block to develop solutions that enable instant communication over language barriers.

In a blog post, the company claimed that their SeamlessM4T model could bring together technology that was previously only available in separate models to give translations between text and speech in about 100 languages as well as full speech-to-speech translation for 35 languages.

CEO Mark Zuckerberg is staking the future of the company on the metaverse, a collection of interconnected virtual worlds, and claims that these technologies will aid users from all over the world in communicating there.

According to the blog post, Meta is releasing the model for general public usage for non-commercial purposes.

The world’s largest social networking site released a flood of largely free AI models this year, including a substantial language model called Llama that directly competes with the proprietary models provided by Google and OpenAI, which is supported by Microsoft.

Because Meta stands to gain more by effectively crowdsourcing the creation of user-facing tools for its social platforms than by charging for access to the models, Zuckerberg argues that Meta benefits from an open AI environment.

Comedian Sarah Silverman and two other authors complained against Meta and OpenAI in July, claiming that the businesses had used their works as training data without their permission.

This data was acquired from 4 million hours of “unprocessed audio sourced from a publicly accessible collection of web data.” However, they did not explicitly mention the specific repository they utilized for this purpose.

A spokeswoman for Meta declined to comment when asked about the origin of the audio data.

The research report claims that text data was obtained from datasets created the previous year that had content taken from Wikipedia and linked websites.

For the most recent tech news, go to or sign up for our newsletter.


Most Popular

Recent Comments