Introducing Falcon ASR

<Hugging Face> Introducing Falcon ASR

Hugging Face has unveiled Falcon-ASR, a speech recognition model tailored for Arabic, boasting 1.6 billion parameters. Developed by the Technology Innovation Institute (TII) in Abu Dhabi, the model supports Arabic, English, French, Spanish, and Portuguese. In benchmark evaluations, Falcon-ASR achieved an impressive average word error rate (WER) of 20.92%, surpassing the current best published result by 2.25 percentage points.

For Emirati dialect, Falcon-ASR showed the lowest WER and character error rate (CER) among the systems it was compared to. The model also transcribes English at a mean WER of 5.74% on public English test sets. Falcon-ASR was trained on a diverse range of dialects, including Emirati, Modern Standard Arabic (MSA), and Gulf dialects, alongside English, to better handle the variations in Arabic speech.

Brief written by urgent.news from Hugging Face's own syndicated text. Machine-written — may contain errors; check the original before relying on it.