Falcon‑ASR & Falcon‑OCR‑Arabic
Arabic AI that listens and reads
Two specialized Falcon models that extend Arabic AI beyond typed text into speech recognition and document understanding.
From spoken words to complex documents
AI interaction does not begin and end with typed text. Falcon-ASR converts spoken language into written text across Arabic and four global languages, while Falcon-OCR-Arabic extracts Arabic text and structured information from images and documents.
Together, they broaden the Falcon ecosystem across two essential ways people interact with information: speech and documents.
Speech recognition built for real-world communication
Falcon-ASR is a compact 1.6-billion-parameter Automatic Speech Recognition model designed to convert spoken language into written text.
- Emirati Arabic
- Modern Standard Arabic
- English
- French
- Spanish
- Portuguese
Compact model. Leading performance.
Falcon-ASR recorded the lowest average word error rate among the systems compared on the Open Universal Arabic ASR Leaderboard evaluation. On TII’s internal Emirati benchmark, it also outperformed larger systems, including a 30-billion-parameter multimodal model.
Note: For WER and CER, lower scores indicate better transcription accuracy.
| Model | Size | Avg WER | Avg CER |
|---|---|---|---|
| Falcon-ASR | 1.6B | 20.92 | 8.79 |
| Audar-ASR-V1-turbo | 2.35B | 23.17 | 9.23 |
| Cohere Transcribe Arabic (07-2026) | 2.0B | 25.87 | 11.80 |
| omniASR LLM 7B | 7.0B | 28.32 | 12.52 |
| Model | Size | WER | CER |
|---|---|---|---|
| Falcon-ASR | 1.6B | 22.73 | 10.19 |
| Qwen3-Omni-30B-A3B-Instruct | 30B (3.0B active) | 26.80 | 12.72 |
| Audar-ASR-V1-Turbo | 2.35B | 27.89 | 13.75 |
| Cohere Transcribe Arabic (07-2026) | 2.0B | 31.05 | 18.07 |
| Qwen3-ASR-1.7B-hf | 2.0B | 31.52 | 13.35 |
| Audar-ASR-V1-Flash | 0.78B | 32.87 | 15.36 |
Built for real audio
Falcon-ASR is designed for real-world audio conditions, including background noise, overlapping speech, room reverberation and telephone-quality sound. It also records word-level timestamps, making it possible to align transcripts precisely with audio and video.
Arabic document understanding at compact scale
Falcon-OCR-Arabic is a 270-million-parameter model built for Arabic optical character recognition and document understanding, extracting text and structured information from images and documents.
High accuracy. Lightweight architecture.
Across 11,974 samples spanning 12 document categories, Falcon-OCR-Arabic achieved 81.9% overall Arabic text accuracy, ranking second among 17 models evaluated and first in four document categories.
Despite its compact size, it outperformed every open-source model in the comparison, including models with significantly larger parameter counts.
The model was evaluated across content including books, official documents, handwritten text, forms, newspapers, receipts, labels, business cards and invoices.
Two Models. Two Modalities. One Expanding Falcon Ecosystem.
Falcon-ASR and Falcon-OCR-Arabic extend Falcon beyond text generation into the information people speak and the documents they use every day, with specialized models designed around real-world Arabic-language needs.