Transparency
What Atqin's recitation tracking actually does, stage by stage — and where the line sits between what's shipped and what's still research.
The pipeline, microphone to screen
Microphone
Your audio is captured in the browser and streamed to Atqin's recognition service over an encrypted connection.
Recognition
A Qur'an-specific speech-recognition model transcribes what it hears, in short passes roughly once a second.
Locator
When you start without picking a passage first, a separate step searches the whole Qur'an for where you are, from audio alone.
Follower
Once your place is found, a different step tracks your position word by word — matched, skipped, repeated, or unclear.
Display
The app renders that evidence: confirmed in green, a possible mistake in the attention colour, everything else neutral.
How recognition actually runs today
Atqin's live recognition uses a Qur'an-specific fine-tuned version of the model below, served through a windowed, near-live path: it re-transcribes your recitation so far roughly once a second, not a persistent low-level cache that only ever looks at new audio. A fully cache-aware, sub-second incremental-streaming architecture exists as an internal research and testing effort, but it is not the path your recitation runs through in production today.
What's available, beta, research, or planned
- Live word-by-word tracking
- Whole-Qur'an locator (Start reciting)
- Confirmed-mistake review
- Tracking-issue reports
- Recognition accuracy across voices, speeds and accents
- Tap-to-relocate when tracking stalls
- Fully cache-aware, sub-second incremental streaming
- Personalised pronunciation practice
- Feedback-strictness modes
- Native iOS/Android apps
What Atqin will and won't claim
Automatic recognition can be wrong
Tracking is evidence-based speech recognition, not a certified transcript — it can miss a real mistake or flag a correct recitation as uncertain.
Uncertainty stays uncertain
When the evidence is ambiguous, a word is labelled uncertain and left there — never silently rounded into a confirmed mistake or a pass.
A timer alone never makes a word red
A word only settles as a possible mistake once the evidence itself resolves that way — never from elapsed time or a single ambiguous pass on its own.
A practice aid, not a teacher
Atqin does not replace a qualified teacher, and does not claim tajweed, makhraj or tashkeel-accuracy analysis beyond what the underlying model genuinely supports today.
Published numbers are dataset-specific
Any accuracy figures Atqin discusses describe performance on a specific held-out test set — not a universal guarantee for every voice, accent or recording condition.
Full detail: Attributions, Privacy, Terms, and Support.