AI & Technology
Your students' recitation graded in seconds
Word-by-word accuracy against the Uthmani Hafs text. Every tajweed rule checked. Every mistake timestamped. No guessing.
Grades in real time
Sessions are processed as they upload. Teachers see results before they finish their morning tea.
Word-level precision
Every word mapped to the Uthmani text. Omissions, substitutions, additions, tashkeel — all flagged with timestamps.
27 tajweed rules
Noon sakinah, madd typology, qalqalah, ghunnah — acoustic analysis, not keyword matching.
Independent benchmark results
Word error rate (WER) on the Hafs EveryAyah dataset. Lower is better.
Dataset: Hafs EveryAyah (114 surahs, full Quran recitation corpus). WER measures word-level transcription accuracy.
How it works
From microphone to score in under a second. The pipeline runs on Modal serverless GPU infrastructure — no cold starts, no queues.
Voice biometrics — who is reciting?
ECAPA-TDNN speaker verification confirms the enrolled student is the one reciting. Enrollment takes 3 Surah recordings, averaged into a phonemic profile. Session verification runs a 3-second sliding window across the audio, checking continuously.
Enrollment
3 Surah recordings averaged into a speaker embedding using ECAPA-TDNN.
Continuous verification
3-second sliding window checks the speaker throughout the session, not just at the start.
Dynamic thresholds
Similarity floor drops when ASR accuracy is high — precise recitation helps clear the check.
Ready to hear the difference?
Start a free trial and run your first session today.
Book a 30-minute demo — we'll walk through the full AI pipeline live for your institution.