The all-in-one audio and timing hub for Qur'anic apps, developers, and researchers. A timestamps visualizer, editing tool and community-verified dataset unifying recitations at scale with word- and letter-level timestamps.
Highlights · Use Cases · Data Access · How it works · Contribute · Roadmap · Acknowledgments · License
-
Unified Qur'anic audio hub: A single consistent schema with comprehensive metadata for reciters and recitations instead of scattered websites, CDN APIs, YouTube playlists, and raw files with different formats.
-
Large-scale, multi-riwayah, multi-style: Full Qur'an coverage across many recitations and hours of audio, spanning mujawwad, murattal, muallim, taraweeh and children repeat styles.
-
Phoneme-based alignment: 20ms phoneme-level precision yields maximum accuracy, eliminates ambiguity at word boundaries and disambiguates tajweed effects where sounds merge across words.
-
Repetition-aware, gap-free timestamps: The pipeline transcribes each silence-based segment independently, so repeated words are detected and timestamped correctly. See the comparison with QUL timestamps.
-
Community-driven validation: No trusting a black-box pipeline. Every stage is automatically checked by dedicated validators and human-correctable through an interactive editing UI. Review flagged errors like missing words or misaligned boundaries, fix them visually, and feed corrections back into the dataset.
-
Submit your own recitations: Add your favorite reciters and different audio sources to the catalog and we handle the processing — typically within a few days.
-
Metadata and versioning: Each recitation is governed by consistent schemas and metadata and versioned with a full history to track segment updates and timestamp corrections over time.
- Verse playback — play or seek any ayah or ayah range straight from the original surah audio.
- Follow-along — word-by-word highlighting synced to the recitation.
- Word study — replay the sound of individual words for learners.
- Tajweed research — measure ghunnah and madd durations from letter timestamps, study cross-word effects and silent-letter interactions, and support tajweed teaching.
- ML research — a large, diverse corpus (reciters, paces, styles, riwayat) for speech recognition, tajweed, recitation start/stop detection, and reciter identification.
Audio, timestamps and metadata ship in two open formats — pick by your use case.
| GitHub Releases | Hugging Face Dataset | |
|---|---|---|
| Best for | Apps, offline use, archives | ML research, analysis, direct audio access |
| Shape | JSON per recitation, in ayah / word / letter tiers | Parquet, one row per ayah |
| Audio | Not bundled — original surah URLs in catalog.json |
Embedded per-ayah clip in every row + original URLs |
| Versioning | Version-pinned snapshots, reproducible | Rolling — always the latest |
| Fetch what you need | Full release or specific reciters | Full dataset or specific reciters, Hugging Face supported live viewer, filtering and querying |
Both formats support both gapless surah and ayah-by-ayah playback. Both ship a single take per full ayah (the first occurrence), so in rare cases where a reciter repeats an ayah fully or partially at the ayah start/end, follow-along highlighting may pause until they move past the repetition (within-ayah repetitions are still preserved). A unified API — which also exposes the full, unfiltered duplicates — is on the roadmap.
| Component | Description |
|---|---|
Quranic Universal Aligner |
Demo running on Hugging Face GPU demonstrating our alignment toolkit, also available via API |
inspector/ |
Entry website for browsing reciters, viewing timestamps interactively and editing alignment results |
| quranic-phonemizer | External package — Qur'an-specific G2P; the foundation that allows phoneme-level alignment |
Visit the website and read the overview info and editing guide to get started in contributing recitations and fixing alignment errors.
Issues and pull requests are welcome. If you've found a bug or have a feature idea, open an issue or jump into the Discord.
To contribute code to the repo directly, fork the repo and see inspector/README.md for setup instructions.
Access
- Unified API/SDKs — typed Python/JS client (
pip/npm) over the published QUA artifacts: fetches and caches only requested data, defaults to latest with optional pinning and offline vendoring, and exposes the schemas for type consistency. Complements the Releases + HF dataset. - Global CDN — mirror all recitations and audio across regions, prewarmed with demand-based routing for low-latency delivery everywhere.
Coverage + Quality
- 100+ recitations — reach 100+ fully aligned and verified recitations.
- Letter-level precision — word and letter timestamps are both high quality; close the few minor systematic and timing differences in letter timestamps that depend on context, tajweed, and reciter.
Generalisation
- Orthography — letter-level timestamps are currently tuned for Uthmani script (DigitalKhatt). Generalise to other scripts where symbols and letter conventions differ, e.g. IndoPak.
- Riwayah — extend beyond Hafs. Each riwayah has its own unique sounds, tajweed, symbols, and ayah orderings, with fewer and less reliable digital assets than Hafs.
- Qur'anic Universal Library (QUL) — Qur'an metadata, the Uthmani script, and the DigitalKhatt font.
- Audio sources — recitations are sourced from QuranicAudio, EveryAyah, MP3Quran, QUL, TVQuran, SurahQuran, and Way2Quran.
The project's own work — timestamps, segmentation, alignment, catalog metadata, and code — is licensed under CC BY 4.0. Recitation recordings remain the property of their reciters and original upstream sources.