Hugging Face introduces Falcon-ASR, 1.6 billion speech model for Arabic
Falcon-ASR is a 1.6 billion parameter speech recognition model for Arabic with a focus on the Emirati dialect; supports English, French, Spanish and Portuguese. Reported average Arabic WER is 20.92% and Emirati WER 22.73%.
What changed
Hugging Face introduced Falcon-ASR, a 1.6 billion parameter ASR trained on Emirati, MSA and other dialects; supports English, French, Spanish, Portuguese and word-level timestamps; demo available.
Why it matters
Developers, researchers and enterprises get an additional Arabic ASR option to evaluate in transcription pipelines and research comparisons.
Who should care
DevelopersResearchersEnterprises
Worth trying
Yes — Demo on Hugging Face; try with your own recordings.