The problem with line-by-line checking
Most apps that listen to your recitation return one of two answers: “correct” or “try again.” That’s not enough when a single mispronounced letter changes a word’s meaning, or when a missed shaddah is the difference between clarity and confusion.
How the verifier works
- You recite a verse aloud.
- Your device captures the audio — on-device when possible.
- Native speech recognition transcribes your recitation.
- Our phonetic matcher aligns your transcription against the expected verse word by word.
- You see a colored replay of the verse:
– Green — matched.
– Amber — close; double-check pronunciation.
– Red — not detected or materially different.
– Grey — you skipped it.
Tap any word to hear the reference reciter pronounce it at your chosen speed.
What it catches
- Skipped words (very common in fast recitation)
- Letter substitutions (seen → ṣād, etc.)
- Wrong vowel lengths (madd errors)
- Misplaced shaddah or sukoon
- Out-of-order verses (a common mutashabihat failure)
- Tajweed rule misses (ghunna, ikhfa, qalqala when tajweed-mode is on)
Privacy — plain language
- Audio is analyzed on-device when your phone supports it.
- When cloud analysis is needed (older devices, some Android variants), audio is encrypted in transit.
- We never use your audio to train our models or any third-party model.
- Recitation audio is automatically deleted from our servers within 30 days.
- You can delete your entire recitation history instantly from Settings → Privacy.
More in AI Ethics.
Accuracy
We’re building a public ground-truth dataset and a recurring evaluation pipeline so we can publish honest, reproducible accuracy numbers — word-level precision and full-line pronunciation scoring, broken down by accent and noise condition. Until that pipeline is live, we won’t post precision figures we can’t show our work on. Watch the changelog for the first published report.
Limitations we’re honest about
- Very noisy environments (traffic, wind) degrade accuracy.
- Strong regional accents may increase false negatives — we’re expanding our acoustic model.
- Children’s voices are handled by a dedicated model; if accuracy drops for your child, please report it so we can improve.
How to try it
Open any memorized verse, tap the microphone, and recite. You’ll see color-coded feedback the moment you finish.
