Meta Unveils New AI Language Model to Preserve Endangered Languages

Meta, a leading technology company, has announced the launch of its new speech-to-text, text-to-speech AI language model. The model can recognize over 4,000 spoken languages and transcribe speeches in over 1,100 languages.

This project aims to preserve languages at risk of disappearing, especially dialects that rely on verbal communication. Also known as Massively Multilingual Speech (MMS), the project combines Meta's work in self-supervised learning, wav2vec 2.0, and a new dataset that provides labelled data for over 1,100 languages and unlabelled data for nearly 4,000 languages. Even languages with no prior speech technology and only a few hundred speakers are included in the project.

Meta has made its models available to the public through the code hosting service, GitHub. It claims that making them open source will assist developers working in various languages in developing new speech applications, such as messaging services built to better understand a wider variety of cultures, or virtual-reality systems that can be used in any language from anywhere.

The Massively Multilingual Speech models outperform existing models and cover ten times the number of languages. Meta hopes that its models and code will encourage others in the research community to build upon its work to preserve the world’s language diversity.

Looking ahead, Meta plans to expand language coverage to support even more languages and address the challenge of dealing with dialects, which is often difficult for existing speech technology.

The model’s goal is to make it easier for people to access information and use devices in the language of their choice. Through its open-source platform, developers can now create speech applications that can be used in any language, regardless of the number of speakers.