Meta Joins the LLM Run

Meta, formerly known as Facebook, has introduced a new Large Language Model (LLM) called Massively Multilingual Speech (MMS). The purpose of this model is to preserve languages worldwide. Unlike other tech giants like Google and Microsoft, Meta's approach to AI is focused on language preservation.

The MMS model expands the capabilities of text-to-speech and speech-to-text technology to over 1,100 languages, which is a significant increase compared to the previous limit of around 100 languages. Additionally, the model can recognize over 4,000 spoken languages, marking a 40-fold increase.

Meta recognizes that many languages around the world are at risk of extinction, and the limitations of current speech recognition and generation technology can contribute to this trend. The company aims to make it easier for people to access information and use tools in their preferred language. The new AI models developed by Meta are designed to address this need.

One of the biggest challenges in creating this model was gathering audio data for thousands of languages. While the largest existing speech dataset covers only around 100 languages, Meta leveraged publicly available audio recordings of New Testament readings in various languages, which provided an average of 32 hours of data per language. By utilizing this extensive dataset, Meta aims to enhance the language capabilities of its AI models and contribute to language preservation efforts.

Overall, Meta's MMS model represents the company's commitment to supporting linguistic diversity and enabling people to access information and tools in their native languages.