TII Releases Falcon-ASR, a 1.6B-Parameter AI Speech Recognition Model for Arabic

On October 7, the Technology Innovation Institute (TII) in Abu Dhabi released Falcon-ASR, a 1.6-billion-parameter speech recognition model for Arabic, with a particular focus on the Emirati dialect. The model also supports English, French, Spanish, and Portuguese.

According to TII, Falcon-ASR achieved an average word error rate (WER) of 20.92% across six Arabic test sets, beating the best published result of 23.17% in the referenced leaderboard snapshot. In an internal Emirati evaluation, the model recorded 22.73% WER and 10.19% character error rate, lower than compared systems; its WER was 4.07 percentage points below Qwen3-Omni, the next best result. The model also provides word-level timestamps linking each transcribed word to its position in the audio.

Falcon-ASR builds on the Falcon3-Audio architecture and training approach. It was trained on Emirati, Modern Standard Arabic, other Gulf and Arabic dialects, and English, with training data including background noise, overlapping speech, music, reverberation, telephony effects, and variations in speed and pitch. Public evaluation data already includes an Emirati subset via Casablanca, and TII supplemented it with internal held-out recordings and human-validated transcripts. On seven English test sets from the Hugging Face Open ASR Leaderboard, the model achieved a mean WER of 5.74%. All five languages share the same weights and output transcripts in the spoken language without requiring a language flag. A Hugging Face demo space is available, with API access and native applications planned.