How Naml works.
How AI follows Quran recitation word by word (and why it runs on your phone)
What a Quran recitation tracker does between your voice and the highlighted word, why it can run on a phone with no server, and what it still gets wrong.
2026-11-19 · draftTracking Tarawih in real time: showing the ayah and translation on screen in your masjid
How a masjid can show the ayah the imam is reciting, the current word and its English on a screen during Tarawih, with no internet and no operator.
2026-11-19 · draftChecking hifz with your phone: a practical memorization test for parents and teachers
A memorization test that asks a random ayah from a range, lets the student continue, and checks the continuation automatically, with the adult still deciding.
2026-11-19 · draftAutomatic tajweed checking: what AI can and can't detect today (shadda, ikhfa, iqlab, madd)
An honest account of automatic tajweed checking on a phone - which mistakes it detects, how often it raises false alarms on correct recitation, and why madd length is not yet one of them.
2026-11-19 · draftTajweed phonemes explained: how we teach a model the difference between madd 2, 4 and 6
Why our Quran speech model writes tajweed phoneme units instead of letters, what the 77 units are, and what a model can and cannot learn about madd length from them.
2026-11-19 · draftTernary (2-bit) speech models: running a Quran ASR model in 17 MB
How we store a streaming speech model's weights as -1, 0 or +1 with one scale per block of 128 inputs, and what that buys on a phone.
2026-11-19 · draftWebAssembly SIMD vs WebGPU for on-device speech recognition: lessons from phones
We expected WebGPU to win for in-browser speech recognition. On a Mali phone our WebGPU step took about 230 ms against 18 ms on the CPU with relaxed SIMD. Here is what we measured and the policy we ship.
2026-11-19 · draftAyah and word timestamps for Quran audio: benchmarking forced alignment models
How accurately can a model say when each ayah starts and ends in a Quran recording? We compared our streaming models with wav2vec2 aligners on 31 reviewed clips.
2026-11-19 · draftBuilding a Quran app with offline speech recognition: privacy, cost and latency
A developer's guide to adding recitation listening to a Quran app without a speech server - what to weigh, what it costs, and the options from free weights to ready source code.
2026-11-19 · draftCommon recitation mistakes and how to catch them: mutashabihat, skipped words and confused letters
The mistakes reciters most often make - similar passages, skipped words and ayahs, confused letters, missing shadda - and how an on-device tracker and tajweed checker catch each one.