Quranic Phonemizer

المحوّل الصوتي القرآني

Turn Quran text into the sounds actually recited. Phoneme by phoneme, shaped by the riwayah, the performance, and where the reader stops—with every tajweed rule traced to the letters and sounds that produce it.

fields
Computational linguisticsArabic phonologyTajweed modellingRule systemsCorpus analysisSpeech technologyAPI design
44 · 49
tajweed rules modelled, Hafs · Warsh
68 · 71
sounds in the inventory, Hafs · Warsh
6 · 1
stop-sign classes honoured, Hafs · Warsh
26 · 57
performance choices modelled, Hafs · Warsh

Version 3.0, 2026-09-04. Counts refresh with every release.

What is broken today

  • Text is not pronunciationWritten Quran text alone does not tell software every sound that surfaces, which letters fall silent, how long a sound lasts, or what changes at a stop. It needs a phonetic model, not another text lookup.
  • One pronunciation cannot stand for every riwayahHafs, Warsh, and the other riwayat have their own sounds, rules, scripts, and authenticated performance choices. Projecting one model onto them erases what should be preserved.
  • The sound changes with the performanceSakt or idraj, tashil or tahqiq, the faces of raa and nasalisation, and the place of stopping all change the phoneme sequence.
  • A colour is not a modelMost tajweed data begins as coloured labels on one text. It cannot generate pronunciation, explain the sound, or follow a different stopping plan.

What a Quranic phonetic model lets you do

Text into phonemesExact sound sequences from Uthmani text. Every consonant, vowel, lengthening, nasal, and release appears in the order it is recited.
  • One sound system per riwayahHafs and Warsh are modelled separately, each with its own script, sound inventory, rules, and authenticated performance choices.
  • Performance and stopping change the soundsChoose sakt or idraj, tashil or tahqiq, a face of raa, or a stopping point. The phoneme sequence is recomputed, not looked up.
  • Every phoneme stays tied to the textEvery sound points back to the Quran letters and vowels that produced it, including sounds shared across letters and letters that fall silent.
  • Tajweed as structured dataEvery rule is named, located, and tied to the sound it changes. A fact in the model, not a colour painted onto text.
  • Search the Quran by ruleEvery verse with an ikhfaa before a given letter, every qalqala that only exists when you stop, every raa that turns heavy: counted across the corpus under the stop plan you choose.
  • Colour any mushaf any wayRule positions come as data. One rule, ten, or none; your palette, your script, your size; the text stays text.
  • Ground truth for modelsSounds and rules for every verse, both riwayat, any stop: what recognition, synthesis, and pronunciation checking train on.

Where we are, where we are heading

Where this goes: one phonetic model per riwayah, and everything that needs to know how the Quran sounds, search, colouring, teaching, timing, models, reads it from there.

  1. More riwayat
  2. Qalun and Shuba
  3. Colouring package and APIRule positions on the mushaf, not a list of rules.
  4. Tajweed searchBy rule, refinement, and stop plan, across the corpus.
  5. 2026Published, with the website
  6. 2026WarshIts own sounds, rules, and script, and the performance choices within it.
  7. 2025HafsSounds, stops, and every rule on its letter.