Quranic Universal Aligner

المحاذي القرآني الشامل

Turn any recitation into timed text. Segments at the reciter’s stops, words and letters timed, shaped how you want.

fields
Speech recognitionForced alignmentSequence alignmentAudio signal processingBenchmarkingAPI design
≈20 s
an hour of audio, aligned
≈180×
faster than real time
12 h
the longest clip, in one go
≈1 min
the same hour without a GPU

Medians of real runs on the Aligner. Every user gets a free GPU allowance each day.

What is broken today

  • Timing rebuilt in every appEvery reader, caption tool, and research pipeline needs audio tied to its text, and each one solves it alone, with its own model and its own failures.
  • Transcription invents wordsA general speech model writes down what it thinks it heard. In the Quran one invented or dropped word is not an error to tolerate.
  • Timed audio for the famous fewA handful of reciters have timings, made once by hand. Your own recording, your teacher’s, your imam’s, has none.
  • Nothing tells you when it is wrongTimings arrive as a finished file. Nothing marks the doubtful parts, so either every one is trusted or all of them are checked by hand.

One recording. What comes back

Any clipA verse, a prayer, a night of tarawih, surahs in any order. Up to twelve hours in one go.
  • Cut at the reciter’s stopsSegments begin and end where the reciter actually paused, never on a fixed clock.
  • Never invents a wordThe audio is matched against the Quran text itself, so what comes back is always what is in the mushaf.
  • Follows repeatsThe reciter goes back over a phrase or a verse; the alignment goes back with him and carries on.
  • Real recitation, real recordingsPhone microphones, room echo, hiss, fast recitation, and the isti’adha, basmala, or du’a in between.
  • Four riwayatHafs, Warsh, Qalun, and Shuba, each aligned against its own text, never projected onto Hafs.
  • Cut it your wayBy verse, by word count, by seconds, with the stops kept at the ends so the meaning stays whole. The audio stays in sync through every change.
  • Words and letters timedEvery word, and every letter inside it, carries its own start and end: captions, highlighting, and learning.
  • Says what it is unsure ofEvery segment carries a score, and anything doubtful is marked for review. Nothing fails quietly.
  • One call from any appAn open API with no key, progress while it runs, and one session reshaped again without aligning again.

Where we are, where we are heading

Where this goes: the engine every app calls when it needs a recitation timed. Any audio, any riwayah, measured in the open.

  1. For teachers and schoolsLetter timing and alignment put plainly, for teaching and learning platforms to build on.
  2. Aligning translationsThe same timeline carried across to translated text.
  3. Captions however you want themRicher shaping and styling, so a timeline leaves ready to use.
  4. Reading what it hearsFirmer confidence, waqf detection, the reciter’s own performance choices, and errors: a word mispronounced or skipped.
  5. Search the Quran by voiceFind likely and similar matches almost instantly, or jump straight to one target. One simple API, with no speech-recognition logic in the app.
  6. Better, faster, on more of what existsQuality, speed, and reach, improved release after release and measured in the open.
  7. nowCleaner cuts, wider reachBetter segmentation, and holding up across every kind of recording.
  8. Sep 2026A public benchmarkHuman-verified recordings anyone can measure a system against: alignment, word timing, and waqf detection.
  9. Sep 2026Four riwayatWarsh, Qalun, and Shuba beside Hafs, each against its own text.
  10. Jun 2026Feeding the libraryReviewed alignments open to every app as Quranic Universal Audio.
  11. Jun 2026Captions you shapeSplit, merge, edit, save, and pick the work up later.
  12. Mar 2026Open to everyoneAny recitation, aligned on the website by anyone, free.