Models
Free and public, never sold. Each model states its licence. Speech recognition comes first; restoration, alignment and data serve it.
- TajweedFormerSpeech recognition
The next recognition model. It was going to be Zipformer Quran 4; it has its own name because the architecture is new.
Evaluation after trainingIn training - Zipformer Quran 3.1 Speech recognition
Writes a recitation down phoneme by phoneme, tajweed included, streaming on a phone with no internet. No language model, so a reciter's mistake is never quietly corrected.
4.13%PER, full 600-clip benchmarkReleased Hugging Face - SAWT v4 Speech restoration
Restores archive recitations to 48 kHz studio sound in the same voice, without moving a phoneme. Each restored window is checked against the input before it is kept.
0.071PER on 120 restored archive recitationsReleased Hugging Face - MFA Quran Hafs Forced alignment
Aligns each ayah to its text phone by phone, so madd, ghunnah and qalqalah can be measured, not only word timestamps.
10–20 msmedian phone-boundary errorReleased Hugging Face - Quranic TTS Speech synthesis
A text-to-speech model for research and teaching. Planned, not in training yet. Its corpus, QuranTTS, is already public.
301 hQuranTTS corpus, publishedPlanned Hugging Face
Datasets
What the models are built and measured on. Contributed voices are never part of a public dataset.
- quran-tajweed-phoneticsTajweed phonetics
Every ayah phonemized, each phone tied to the tajweed ruling behind it and its citation in the classical books.
208rulings reviewed by expertsPublic Hugging Face - quranic-asr-benchmarkHeld-out test sets
Studio reciters, a QUL reciter and real phone recordings, none of them trained on.
600clips in 3 setsPublic Hugging Face - QuranTTSSpeech corpus
An ear-verified recitation corpus for TTS and restoration: 16 reciters, 48 kHz, 24-bit.
301 h64,721 clipsPublic Hugging Face
Quranic ASR leaderboard
Any model, scored on the same held-out sets. Submissions run on Hugging Face.