hetchyy/quranic-universal-ayahs

Qur'anic Universal Ayahs

Qur'anic Universal Audio (QUA) is a project that unifies recitations on the internet and generates timing data using forced alignment — community-verified results and constantly expanding dataset.

This dataset pairs ayah by ayah audio with word-level timestamps, letter timestamps, and waqf-aware segment data. Repeated words are preserved in text_uthmani and word_timestamps, so the row reflects what the reciter actually recited rather than a plain copy of canonical ayah text.

Tip: Click the three dots (···) at the top right and toggle Notifications to get updates whenever recitations are added or refreshed.

Dataset

Mushaf configs contain one row per ayah. Each row includes embedded ayah audio, recited Uthmani text, word timestamps, letter timestamps, waqf-aware segments, and source-audio mappings.

The mushafs subset is the dataset catalog index. It has one row per published mushaf, joined with reciter names, riwayah/style/channel labels, audio metadata, and coverage.

Remaining subsets are grouped by riwayah, and splits are specific mushafs.

Mushaf Schema

Column	Type	Notes
audio	Audio	Ayah MP3 clip.
surah	int32	Surah number, 1-114.
ayah	int32	Ayah number within the surah.
duration_ms	int32	Clip duration.
text_uthmani	string	Recited Uthmani text (normalized)
segments	[[int,int,int,int]]	Waqf/pause-aware regions: [word_from, word_to, start_ms, end_ms].
word_timestamps	[[int,int,int]]	[word_idx, start_ms, end_ms]; word indices are 1-based.
letter_timestamps	struct of lists	word_idx, char, start_ms, end_ms. char is one token from a fixed 42-token alphabet — see letter_vocab_hafs_qpc.csv in this repo.
source_url	string	Original chapter or ayah audio URL.
source_offset_ms	int32	Clip start inside source_url.

All row timestamps are relative to the ayah clip. Use source_offset_ms + timestamp_ms when mapping a row back to its source audio.

Catalog Schema

One row per published mushaf.

Column	Type	Notes
slug	string	Stable mushaf identifier (the per-riwayah split name).
reciter_id	string	Stable reciter identifier (shared across that reciter's mushafs).
name_en, name_ar	string	Reciter display name (English / Arabic).
country	string	Reciter country, ISO 3166-1 alpha-2 (e.g. SA).
riwayah	string	Riwayah name (e.g. Hafs).
style	string	Recitation style (e.g. Murattal, Mujawwad).
recording_context	string	Recitation context (e.g. Studio, Taraweeh).
recording_year	int32	Year recorded, when known.
channel	string	Distribution channel name.
audio_category	string	How source audio is segmented: by_surah | by_ayah.
chapter_count	int32	Number of chapters in the mushaf.
codec	string	Audio codec.
container	string	File container.
sample_rate_hz	int32	Sample rate.
channels	int32	1 = mono, 2 = stereo.
bitrate_mode	string	cbr | vbr | mixed | unknown. mixed = chapters differ.
bitrate_kbps_nominal	int32	Nominal bitrate; null when bitrate_mode is mixed.
total_duration_hours	float32	Total mushaf audio duration, in hours (1 dp).
published_at	string	When the mushaf first entered the dataset, ISO-8601.
updated_at	string	When the mushaf was last published/refreshed, ISO-8601.

Usage

from datasets import load_dataset

ds = load_dataset(
    "hetchyy/quranic-universal-ayahs",
    "hafs_an_asim",
    split="khalifa_al_tunaiji_tarteel",
)

row = ds[0]
print(row["surah"], row["ayah"])
print(row["word_timestamps"])
print(row["segments"])

catalog = load_dataset(
    "hetchyy/quranic-universal-ayahs",
    "mushafs",
    split="all",
)

print(catalog[0]["name_en"], catalog[0]["riwayah"])

Notes

• Best for quick access to verse audio and timestamps together, ayah-by-ayah playback, and ML research.
• Also see GitHub Releases, a parallel format using JSON files with versioning and checksums, suitable for offline usage.
• Recitation audio is not relicensed, and remains the property of upstream sources/reciters.

License

CC BY 4.0.