Nabr
Acoustic model for Quranic recitation: ASR and Tajweed-aware mispronunciation detection.
About Nabr
Specialized acoustic model for Quranic recitation analysis and mispronunciation detection. Not a general speech-to-text model adapted for Arabic, but trained on the actual phonetics of Tilawa. Analyzes 8 acoustic properties per sound (Madd, Ghunnah, Qalqalah, Tafkhim). Catches errors a human Quran teacher would catch. Character error rate: 0.59%, Word error rate: 2.26%. Real-time processing; already in production in the Eqraa Quran app.
Capabilities
Where it runs
Nabr powers real-time recitation correction in Qurani.ai, Misraj's API for Quranic data. Its recitation-correction (QRC) API uses Nabr to flag Tajweed errors at the level of individual sounds, including Madd, Ghunnah, Qalqalah and Tafkhim, and it is live in the Eqraa Quran app.
Research
The work behind Nabr is described in Annotating Real-World Quranic Recitations for Mispronunciation Detection and Diagnosis, a Misraj paper that introduces a Tajweed-aware corpus for Quranic mispronunciation detection. The paper is under review, so its full results are not public yet.
Deployment
Nabr is available through the Qurani.ai API and the Misraj cloud, or on-premises.
At a glance
- Type
- Speech
- License
- Commercial
- Deployment
- Cloud APIOn-premises
- Research track
- Islamic text & acoustic intelligence
- Links
- Try it
Papers
Islamic text & acoustic intelligence
Annotating Real-World Quranic Recitations for Mispronunciation Detection and Diagnosis
Let's talk about what you're trying to build.
Every good solution starts with a clear question. Ask us, and our team is with you from question to solution.