TII introduces Falcon-ASR for Arabic speech recognition
TII's new speech recognition model focuses on Emirati Arabic, supports four other languages and is available through an online demo.
Abu Dhabi's Technology Innovation Institute has introduced Falcon-ASR, a speech recognition model focused on Arabic and the Emirati dialect. The model has 1.6 billion parameters and also transcribes English, French, Spanish and Portuguese.
TII said its testing produced a 20.92% average word error rate across six Arabic datasets. The strongest published comparison in its selected leaderboard snapshot scored 23.17%, with lower scores indicating fewer errors. Falcon-ASR recorded a 22.73% word error rate on TII's internal Emirati evaluation.
The project addresses regional speech that TII says is underrepresented in AI training data. Its transcripts include timing information for individual words, supporting subtitles and searchable recordings. The institute introduced the model alongside separate systems for Emirati language generation and Arabic document processing.
An online demonstration is available for testing recordings. TII said API access and native applications are planned, but its announcement gave no release dates for those options.
The day's biggest AI news in one short email, every morning. Plus an instant heads-up when a reset or a big release lands.