Overview
[AI Summary]: Meta AI has released Omnilingual ASR, an open-source automatic speech recognition system supporting over 1,600 languages, including 500+ low-resource languages previously unsupported by existing ASR systems. The technology achieves high accuracy using large-scale models (up to 7B parameters) and features in-context learning capabilities inspired by LLMs, enabling zero-shot learning for new languages with just a few samples. The project includes both the models and the Omnilingual ASR Corpus dataset, all released under Apache 2.0 license to improve digital accessibility for global language communities.
- Developer: Meta AI / Facebook Research
- License: Apache 2.0
- Platform: Cloud (7B model) and on-device (300M lightweight model)
- Languages Supported: 1,600+
- Key Features: Self-supervised learning, encoder-decoder architecture, in-context learning
- Online Demo: Omnilingual ASR Media Transcription - a Hugging Face Space by facebook
- GitHub: GitHub - facebookresearch/omnilingual-asr: Omnilingual ASR Open-Source Multilingual SpeechRecognition for 1600+ Languages
- Research Paper: https://ai.meta.com/research/publications/omnilingual-asr-open-source-multilingual-speech-recognition-for-1600-languages