Meta’s Open Source AI Project Can Identify 4,000+ Spoken Languages

Meta ai

Meta, the parent company of Facebook, has unveiled its Massively Multilingual Speech (MMS) project, introducing advanced speech-to-text, text-to-speech capabilities, and more for over 1,100 languages. This leap forward in AI technology aims to democratize access to information across the globe, even for languages that are at risk of vanishing.

Meta’s MMS project incorporates their self-supervised learning work, wav2vec 2.0, and a new dataset that provides data for a wide range of languages, some of which have been unrepresented in prior speech technology. As a result, MMS models have proven to outperform existing models, supporting ten times as many languages.

Key to this achievement is a vast audio dataset assembled from readings of the New Testament translated into over 1,100 languages, as well as other Christian religious readings, expanding the language coverage to over a whopping 4,000 languages.

Despite its religious origins, Meta’s analysis reveals that the resulting speech models perform equally well for male and female voices and are not overly biased toward religious language.

The MMS project has shown promising results, offering competitive performance in comparison with other models like OpenAI’s Whisper, while covering significantly more languages. The project also supports language identification for over 4,000 languages, with encouraging results on the VoxLingua-107 benchmark.

Meta says it is publicly sharing its models and code, hoping that this contribution will aid in preserving the world’s rich linguistic diversity. The company is also working towards developing a single model capable of performing multiple speech tasks for all languages, further broadening its accessibility potential.

As part of this project, Meta remains mindful of the limitations and potential challenges of AI technology, including the risk of mistranscribing words or phrases which could result in offensive or inaccurate language.

This is a pretty impressive AI project and the fact it’s open source is great news, as it ensures anyone can use it to preserve their rare languages that might be on the verge of disappearing (like Klingon). AI is going to change humanity, folks.