Research
Notes on how the models are built and measured, written by the people who built them.
The data, in numbers
Read how it was built
- Recitation in the corpus
- 59,000+ h
- Ayat phonemized
- 6,236
- Rulings reviewed by experts
- 208
- QuranTTS, public
- 301 h
- Benchmark clips, held out
- 600
- Nov 2026 TajweedFormer V4, and an SDK to put it in your app Our recitation model now hears every sound with its timing, and lengthening, gemination, ghunna and qalqalah beside it, inside the app. Free SDKs for Android, Apple, the web, Windows and Linux. ASR
- Oct 2026 RecFlow: 884,821 recitations from our own tajweed TTS Our own tajweed TTS recites from the Quranic text, holds a madd for the counts it is given, and writes the training data that teaches our recitation model to hear mistakes. TTS
- Oct 2026 Every phone of the Quran, with the rule behind it quran-tajweed-phonetics: all 6,236 ayat as 522,475 phones, each tied to its tajweed ruling and its source in the classical books, reviewed by scholars. Data
- Oct 2026 SAWT V6.1 and V6.2: what changed after release The pause gate stays out of speech, digital silence stays silent, and Pro has new weights with lower phoneme error on the hardest recordings. Restoration
- Oct 2026 SAWT V6: the pauses fall silent SAWT V6 Pro and V6 Flash restore archive recitation to 48 kHz with silent pauses and the melody kept, ahead of every open restorer we measured. Restoration
- Sep 2026 Proving the potential of ternary quantization in ASR An internal experiment: Zipformer Quran 3.1 at −1, 0 and +1 reaches 3.34% phoneme error, but it is overfit to the canonical text. Not for feedback yet; ready to build a canonicalizer on. ASR
- Sep 2026 The data behind the models: a 59,000-hour corpus, and what the models took from it Quran Lab holds more than 59,000 hours of recitation. No model trains on all of it; this is how the selections for Zipformer Quran 3 and 3.1 were made. Data
- Sep 2026 Zipformer Quran 3.1: it now writes what was said 4.13% phoneme error on all 600 benchmark clips, madd held as it was recited, and the Neural Engine. ASR
- Sep 2026 SAWT v4: 48 kHz restoration of archive recitations Bringing archive recitations back to studio quality with a 372M flow-matching generator that renders texture but never decides what was said. Restoration
- Sep 2026 Zipformer Quran 3: closing the emphatic-letter gap How ص ض ط ظ went from three times the error of plain letters to parity on every held-out set. ASR