Falcon-ASR, a 1.6 billion parameter model from Abu Dhabi's Technology Innovation Institute, transcribes Arabic with lower error rates than previous systems on benchmark tests. It handles dialects, code-switching, and noisy recordings.
What changed
Falcon-ASR is available via a Hugging Face demo. It transcribes Arabic, English, French, Spanish and Portuguese using shared model weights, without requiring users to specify the language. Word-level timestamps link each transcribed word to its audio position. API access and native apps are planned but not yet released.
Original source
This report summarises the source below. Analysis is labelled separately; product and research claims remain attributed to their source.
Read the original at Hugging Face