quranlab/quran-audio

QuranLab — Quran Recitation Audio

A verse-aligned, honestly-licensed reference layer for Quran recitation audio: a canonical reciter taxonomy of 245 reciters + a per-ayah reference manifest (47 recordings, everyayah) keyed on the canonical 6,236-ayah verse_key (Hafs ʿan ʿAsim) + a per-surah reference layer (285 recordings, mp3quran, spanning 15 qiraat/riwayat) + word-precise CC-BY timing for 44 recordings (11 from cpfair/quran-align + 33 we generate ourselves via permissive forced alignment — see below).

⚠️ This dataset hosts NO audio bytes — by design and by law. A Quran recitation is a separately-copyrighted sound recording (the reciter's performance + the producer's master), independent of the Quran text (which is itself non-copyrightable). None of the surveyed sources grants an open license to redistribute the recordings, and DMCA §512(c) safe harbor does not cover files an operator uploads itself — so re-hosting commercial recitations would be willful infringement, not a takedown-curable risk. We therefore publish only: (1) the reciter taxonomy, (2) per-ayah audio_url references to the original CDN (you fetch the audio yourself, under the source's terms), and (3) CC-BY-4.0 word timing — both from cpfair/quran-align and generated by us via a fully permissive forced-alignment toolchain (Apache-2.0 acoustic model + torchaudio), released CC-BY-4.0. See LICENSES.md.

Companion to the text corpus quranlab/quran — join on verse_key to pull Arabic text / translations / tafsir, and on (verse_key, word_position) to the morphology-ar word layer (word_position is the 1-based word index in each segments entry).

Configs

• manifest (default) — slim cross-reciter index: every (recitation × ayah) with its audio_url. Best for "give me ayah 2:255 across all reciters".
• reciters — the canonical reciter taxonomy (245 reciters: id, names ar/en, nationality, era, tier, sources, riwayat & styles available). The piece no existing audio set provides cleanly.
• surah-manifest — the per-surah reference layer: every (mp3quran recording × surah) with a whole-surah surah_audio_url. 285 recordings spanning 15 riwayat/turuq — the breadth layer (hundreds of reciters, all major readings).
• <recitation_id> — one config per per-ayah everyayah recording (47); the 44 with timing also carry a segments word-timing column + duration_ms.

from datasets import load_dataset
m   = load_dataset("quranlab/quran-audio")                                  # per-ayah manifest (default)
rc  = load_dataset("quranlab/quran-audio", "reciters",       split="train") # canonical taxonomy
sm  = load_dataset("quranlab/quran-audio", "surah-manifest", split="train") # per-surah breadth
hus = load_dataset("quranlab/quran-audio", "husary",         split="train") # one recording + word timing

Schema (per-recitation config)

verse_key · surah · ayah · ayah_global (1–6236) · recitation_id · reciter_id · name_en · name_ar · riwayah · style (murattal/mujawwad/muallim) · enumeration · alignment_basis · numbering_divergence_risk · tier · audio_url (external; no bytes) · granularity · bitrate · bitrate_variants · has_word_timing · source · source_url · license · attribution · commercial_ok. Timed recordings additionally carry segments (list<{word_start, word_end, word_position, start_ms, end_ms}> — word_start 0-based and word_end exclusive per cpfair; word_position is 1-based and joins to morphology-ar.word_index; start_ms/end_ms are offsets within the ayah audio) · duration_ms · timing_source (cpfair/quran-align or quranlab-forced-alignment) · timing_license (CC-BY-4.0 either way) · timing_attribution. Note: forced-alignment timing has ~1% artifacts (zero-duration or merged/out-of-order spans), kept verbatim.

How we generate our own timing

For recordings cpfair doesn't cover, we generate word timing ourselves and release it CC-BY-4.0, using a fully permissive toolchain: forced alignment (torchaudio.functional.forced_align, BSD) of each ayah's verbatim Tanzil Uthmani text against that ayah's everyayah audio, with the Apache-2.0 acoustic model jonatasgrosman/wav2vec2-large-xlsr-53-arabic. Audio is fetched transiently and discarded (no bytes hosted). Validated against cpfair (Husary): mean |Δ word-start| ≈ 114 ms, on par with cpfair's own ~73 ± 139 ms. Reproducible via python -m src.audio.align <reciter_id>. Coverage is marked per row by has_word_timing — for recordings whose alignment is still in progress this may be partial (some ayahs timed, the rest not yet); a later rebuild completes them.

Schema (surah-manifest, per-surah)

surah (1–114) · surah_recitation_id · reciter_id · name_en · name_ar · riwayah · style · numbering_divergence_risk · granularity (per-surah) · surah_audio_url (external; no bytes) · moshaf_label_ar · mp3quran_reciter_id · source · source_url · license · attribution · commercial_ok. No verse_key — a surah is a surah across qiraat, so per-surah rows need no ayah-level alignment; riwayah is recorded for filtering.

Alignment & riwayah

Per-ayah audio maps deterministically to verse_key (everyayah SSSAAA.mp3). Non-Hafs recordings (Warsh) are separate configs and flagged numbering_divergence_risk=true: Warsh follows the Madinan ayah division (~6214/6217 verses) which diverges from the Kufan/Hafs 6,236 count (e.g. Hafs 2:58 = Warsh 2:57). everyayah serves its Warsh sets on the Kufan 6,236 grid (alignment_basis="mapped-to-kufan-6236") so they can be joined on verse_key, but the ~22 divergent loci are cut on Kufan, not Warsh, boundaries — treat with care.

Reciter taxonomy

245 reciters across tier-1 (globally essential), tier-2 (very common), and tier-3 (notable). Each reciter row carries sources (everyayah / mp3quran). Scope note: unlike the text corpus quranlab/quran (which is gated to Ahl-al-Sunnah translations/tafsir), recitation audio is the canonical Quran text recited in the established qiraat, so no theological gate is applied to reciters. Per-ayah recordings (47):

recitation (config)	reciter	riwayah	style	bitrate	word timing
abdul-basit	Abdul Basit Abdus Samad	hafs-asim	murattal	64kbps	✅ cpfair (CC-BY)
abdul-basit-mujawwad	Abdul Basit Abdus Samad	hafs-asim	mujawwad	128kbps	✅ cpfair (CC-BY)
abdul-basit-warsh	Abdul Basit Abdus Samad	warsh-nafi	murattal	128kbps	—
abdul-rahman-al-sudais	Abdul Rahman Al-Sudais	hafs-asim	murattal	64kbps	✅ quranlab (CC-BY)
ali-al-hudhaify	Ali Al-Hudhaify	hafs-asim	murattal	128kbps	✅ quranlab (CC-BY)
husary	Mahmoud Khalil Al-Husary	hafs-asim	murattal	128kbps	✅ cpfair (CC-BY)
husary-muallim	Mahmoud Khalil Al-Husary	hafs-asim	muallim	128kbps	✅ cpfair (CC-BY)
husary-mujawwad	Mahmoud Khalil Al-Husary	hafs-asim	mujawwad	128kbps	✅ quranlab (CC-BY)
maher-al-muaiqly	Maher Al-Muaiqly	hafs-asim	murattal	128kbps	✅ quranlab (CC-BY)
mahmoud-ali-al-banna	Mahmoud Ali Al-Banna	hafs-asim	murattal	32kbps	✅ quranlab (CC-BY)
minshawi	Mohamed Siddiq El-Minshawi	hafs-asim	murattal	128kbps	✅ cpfair (CC-BY)
minshawi-mujawwad	Mohamed Siddiq El-Minshawi	hafs-asim	mujawwad	64kbps	✅ cpfair (CC-BY)
mishary-alafasy	Mishary Rashid Alafasy	hafs-asim	murattal	128kbps	✅ cpfair (CC-BY)
muhammad-ayyub	Muhammad Ayyub	hafs-asim	murattal	128kbps	✅ quranlab (CC-BY)
mustafa-ismail	Mustafa Ismail	hafs-asim	murattal	48kbps	✅ quranlab (CC-BY)
saad-al-ghamdi	Saad Al-Ghamdi	hafs-asim	murattal	40kbps	✅ quranlab (CC-BY)
saud-al-shuraim	Saud Al-Shuraim	hafs-asim	murattal	128kbps	✅ cpfair (CC-BY)
abdullah-al-juhani	Abdullah Awad Al-Juhany	hafs-asim	murattal	128kbps	✅ quranlab (CC-BY)
abdullah-basfar	Abdullah Basfar	hafs-asim	murattal	64kbps	✅ quranlab (CC-BY)
abu-bakr-al-shatri	Abu Bakr Al-Shatri	hafs-asim	murattal	128kbps	✅ cpfair (CC-BY)
ahmad-al-ajmi	Ahmad Al-Ajmi	hafs-asim	murattal	128kbps	✅ quranlab (CC-BY)
ali-jaber	Ali Abdullah Jaber	hafs-asim	murattal	64kbps	✅ quranlab (CC-BY)
ayman-suwaid	Ayman Rushdi Suwaid	hafs-asim	murattal	64kbps	✅ quranlab (CC-BY)
fares-abbad	Fares Abbad	hafs-asim	murattal	64kbps	✅ quranlab (CC-BY)
hani-ar-rifai	Hani Ar-Rifai	hafs-asim	murattal	64kbps	✅ cpfair (CC-BY)
khalifa-al-tunaiji	Khalifa Al-Tunaiji	hafs-asim	murattal	64kbps	✅ quranlab (CC-BY)
mohamed-al-tablawi	Mohammad Al-Tablawi	hafs-asim	murattal	128kbps	✅ cpfair (CC-BY)
muhammad-jibreel	Muhammad Jibreel	hafs-asim	murattal	128kbps	✅ quranlab (CC-BY)
nasser-al-qatami	Nasser Al-Qatami	hafs-asim	murattal	128kbps	✅ quranlab (CC-BY)
yasser-al-dosari	Yasser Al-Dosari	hafs-asim	murattal	128kbps	✅ quranlab (CC-BY)
abdullah-matroud	Abdullah Al-Matrood	hafs-asim	murattal	128kbps	✅ quranlab (CC-BY)
ahmed-neana	Ahmed Neana	hafs-asim	murattal	128kbps	✅ quranlab (CC-BY)
akram-al-alaqimy	Akram Al-Alaqimy	hafs-asim	murattal	128kbps	✅ quranlab (CC-BY)
ali-hajjaj-al-suesy	Ali Hajjaj Al-Suesy	hafs-asim	murattal	128kbps	✅ quranlab (CC-BY)
aziz-alili	Aziz Alili	hafs-asim	murattal	128kbps	✅ quranlab (CC-BY)
ibrahim-al-akhdar	Ibrahim Al-Akhdar	hafs-asim	murattal	32kbps	✅ quranlab (CC-BY)
ibrahim-al-dosary-warsh	Ibrahim Al-Dosary	warsh-nafi	murattal	128kbps	—
karim-mansoori	Karim Mansoori	hafs-asim	murattal	40kbps	✅ quranlab (CC-BY)
khalid-al-qahtani	Khalid Abdullah Al-Qahtani	hafs-asim	murattal	192kbps	✅ quranlab (CC-BY)
muhammad-abdulkareem	Muhammad Abdul Kareem	hafs-asim	murattal	128kbps	✅ quranlab (CC-BY)
muhsin-al-qasim	Abdul Muhsin Al-Qasim	hafs-asim	murattal	192kbps	✅ quranlab (CC-BY)
parhizgar	Shahriar Parhizgar	hafs-asim	murattal	48kbps	✅ quranlab (CC-BY)
sahl-yassin	Sahl Yasin	hafs-asim	murattal	128kbps	✅ quranlab (CC-BY)
salah-al-budair	Salah Al-Budair	hafs-asim	murattal	128kbps	✅ quranlab (CC-BY)
salah-bukhatir	Salah Abdul Rahman Bukhatir	hafs-asim	murattal	128kbps	✅ quranlab (CC-BY)
yaser-salamah	Yaser Salamah	hafs-asim	murattal	128kbps	✅ quranlab (CC-BY)
yassin-al-jazaery-warsh	Yassin Al-Jazaery	warsh-nafi	murattal	64kbps	—

Removal / takedown

Reciters, producers, or rights holders may request removal of any reference/metadata rows via this dataset's Community (Discussions) tab; they will be removed promptly. (No audio is hosted to take down.)

Roadmap

• Phase 1 (done) — canonical reciter taxonomy + everyayah per-ayah reference manifest (47 recordings, incl. 3 Warsh on the Kufan grid) + cpfair CC-BY word timing (11).
• Phase 2 (done) — the per-surah mp3quran reference layer: 285 recordings across 15 qiraat/riwayat, expanding the taxonomy to 245 reciters (each tagged with its sources). Reference-only, same posture.
• Phase 3 (this release) — our own CC-BY word timing for 33 more recordings, generated via permissive forced alignment (Apache-2.0 model + torchaudio), bringing timed recordings to
  44. (cpfair-locked word timing was the gap; quran.com API is ToS-locked, QUL is auth-walled, everyayah timings are CC-BY-NC — so we made our own, openly.)
• Phase 4 (gated) — an embeddable-audio config, only for recordings with written, sublicensable open-license / true-PD clearance (currently none).