README.md

---
language:
- ar
license: apache-2.0
size_categories:
- 100K<n<1M
task_categories:
- automatic-speech-recognition
- text-to-speech
- audio-classification
pretty_name: Quranic Recitation Dataset (Word-by-Word)
tags:
- quran
- audio
- speech
- religion
- arabic
- word-by-word
- recitation
- opus
- protobuf
---

<p align="center">
  <img src="./banner.png" alt="Quranic Recitation Dataset Banner" width="100%">
</p>

<p align="center">
  <img src="https://img.shields.io/badge/Audio%20Format-Opus%20(16kHz,%2032kbps)-emerald?style=for-the-badge&logo=opus" alt="Opus Format">
  <img src="https://img.shields.io/badge/Total%20Reciters-34-blue?style=for-the-badge" alt="Total Reciters">
  <img src="https://img.shields.io/badge/Timing%20Format-Protocol%20Buffers-violet?style=for-the-badge" alt="Protobuf">
  <img src="https://img.shields.io/badge/CDN-Cloudflare%20Edge-orange?style=for-the-badge&logo=cloudflare" alt="Cloudflare CDN">
  <img src="https://img.shields.io/badge/License-Apache%202.0-green?style=for-the-badge" alt="License">
</p>

---

## 🌟 Overview

**Quranic Recitation Dataset (Word-by-Word Sync)** is a highly optimized, production-ready dataset containing high-quality audio recitations of the Holy Quran synchronized at the **word-by-word** level. 

This dataset features **34 world-renowned reciters**, with every Surah (114 chapters) mapped precisely to millisecond-accurate word timestamps. It is designed for modern Islamic mobile and web applications — served via a **Cloudflare Edge CDN** with native range-request support for instant audio streaming and seeking.

---

## 🚀 Key Features

* **34 Reciters × 114 Surahs** — totaling **3,876 audio files** with matching timing data.
* **Optimized Audio** — Mono-channel, 16kHz, 32kbps `.opus` format. Exceptional clarity at ultra-low bandwidth.
* **Per-Surah Architecture** — Each surah is a single `.opus` file. Seek to any ayah using the timing data — no need to download individual ayah files.
* **Protocol Buffer Timing** — Binary `.pb` timing files are ~33% smaller than JSON and parse instantly in any language.
* **Cloudflare Edge CDN** — Files are served from `cdn.mualim.app` with edge caching, CORS, and native HTTP `206 Partial Content` for mobile audio seeking.
* **Word-Level Accuracy** — Timestamps map to standard Uthmani script word indices, enabling karaoke-style highlighting.

---

## 📡 CDN Access

This dataset is served via a **Cloudflare Worker** that proxies HuggingFace with aggressive edge caching. You never need to download the 17GB repository locally.

### Base URL
```
https://cdn.mualim.app
```

### URL Format
```
{BASE_URL}/{slug}/{surah_number}.{opus|pb}
```

### Quick Examples
```bash
# Audio — Mishari Rashid Al-Afasy, Surah Al-Fatihah
curl https://cdn.mualim.app/mishari-rashid-al-afasy-murattal-hafs/001.opus

# Timing — Same reciter, same surah (Protocol Buffer)
curl https://cdn.mualim.app/mishari-rashid-al-afasy-murattal-hafs/001.pb

# Range Request (for mobile seeking)
curl -H "Range: bytes=0-1024" \
  https://cdn.mualim.app/mishari-rashid-al-afasy-murattal-hafs/001.opus
# → 206 Partial Content ✅
```

### CDN Features
| Feature | Details |
|---|---|
| **Edge Caching** | `max-age=31536000, immutable` (1 year) |
| **Range Requests** | Native `206 Partial Content` support |
| **CORS** | `Access-Control-Allow-Origin: *` |
| **Stale Fallback** | `stale-if-error=604800` (7 days) |
| **Fallback Origin** | Direct HuggingFace if CDN cache misses |

### Direct HuggingFace Fallback
If the CDN is unavailable, you can fetch directly from HuggingFace:
```
https://huggingface.co/datasets/zaibihassan/Quranic-Recitation-Data/resolve/main/{Folder Name}/{Surah}/{File}
```

> **Note:** Folder names on HuggingFace contain spaces and special characters (e.g., `Mishari Rashid Al Afasy (Murattal – Hafs)`). The CDN slug mapping handles this automatically.

---

## 📂 Dataset Structure

The dataset is organized by **Reciter Folder** → **Surah Number (001–114)**.

Each surah folder contains two files:
1. `{surah}.opus` — Per-surah audio file (full surah in one file)
2. `{surah}.pb` — Protocol Buffer word-by-word timing data

```
Quranic-Recitation-Data/
├── Abdul Basit Abdul Samad (Mujawwad – Hafs)/
│   ├── 001/
│   │   ├── 001.opus    (Full Surah Al-Fatihah audio)
│   │   └── 001.pb      (Word timestamps for all ayahs)
│   ├── 002/
│   │   ├── 002.opus
│   │   └── 002.pb
│   └── ...
├── Mishari Rashid Al Afasy (Murattal – Hafs)/
├── Alijon Qari/
└── ...
```

---

## ⏱️ Timing Data Schema (Protocol Buffers)

The `.pb` timing files map every verse in a surah to word-level timestamps.

### Protocol Buffer Schema (`recitation.proto`)
```protobuf
syntax = "proto3";
package recitation;

message WordSegment {
  int32 word_index_0_based = 1;
  int32 word_index_1_based = 2;
  int32 timestamp_from = 3;      // milliseconds from start of surah
  int32 timestamp_to = 4;        // milliseconds from start of surah
}

message VerseSegments {
  repeated WordSegment segments = 1;
}

message SurahTimestamps {
  map<string, VerseSegments> verses = 1;  // key = "surah:ayah" (e.g., "1:1")
}
```

### How It Works

Each `.pb` file contains a `SurahTimestamps` message with all verses of that surah. The timestamps are **relative to the start of the surah audio file**, so you can:

1. Load the surah `.opus` audio
2. Decode the `.pb` timing data
3. To play a specific ayah, seek the audio to `verses["1:3"].segments[0].timestamp_from` milliseconds
4. Highlight words in real-time as `currentTimeMs` falls between each segment's `timestamp_from` and `timestamp_to`

---

## 🔗 CDN Slug Map

The CDN uses URL-friendly slugs that map to the actual folder names on HuggingFace:

| # | CDN Slug | Folder Name |
|---|---|---|
| 1 | `abdul-basit-abdul-samad-mujawwad-hafs` | Abdul Basit Abdul Samad (Mujawwad – Hafs) |
| 2 | `abdul-basit-abdul-samad-murattal-hafs` | Abdul Basit Abdul Samad (Murattal – Hafs) |
| 3 | `abdul-rahman-al-sudais-murattal-hafs` | Abdul Rahman Al Sudais (Murattal – Hafs) |
| 4 | `abdullah-hamad-abu-sharida` | Abdullah Hamad Abu Sharida |
| 5 | `abu-bakr-al-shatri-murattal-hafs` | Abu Bakr Al Shatri (Murattal – Hafs) |
| 6 | `ahmed-ibn-ali-al-ajmy` | Ahmed ibn Ali Al Ajmy |
| 7 | `alijon-qari` | Alijon Qari |
| 8 | `alnufais` | Alnufais |
| 9 | `alzain-mohammad-ahmad` | Alzain Mohammad Ahmad |
| 10 | `ayman-suwaid-muallim` | Ayman Suwaid (Muallim) |
| 11 | `badr-al-turki` | Badr Al Turki |
| 12 | `bandar-baleela` | Bandar Baleela |
| 13 | `fatih-seferagic` | Fatih Seferagic |
| 14 | `hani-ar-rifai-murattal-hafs` | Hani Ar Rifai (Murattal – Hafs) |
| 15 | `idris-abkar` | Idris Abkar |
| 16 | `khalid-al-jalil-murattal-hafs` | Khalid Al Jalil (Murattal – Hafs) |
| 17 | `khalifa-al-tunaiji-murattal-hafs` | Khalifa Al Tunaiji (Murattal – Hafs) |
| 18 | `maher-al-muaiqly-murattal-hafs` | Maher Al Muaiqly (Murattal – Hafs) |
| 19 | `mahmoud-khalil-al-husary-mujawwad-hafs` | Mahmoud Khalil Al Husary (Mujawwad – Hafs) |
| 20 | `mahmoud-khalil-al-husary-murattal-hafs` | Mahmoud Khalil Al Husary (Murattal – Hafs) |
| 21 | `mahmoud-khalil-al-husary-murattal-hafs-v2` | Mahmoud Khalil Al Husary (Murattal – Hafs – V2) |
| 22 | `mishari-rashid-al-afasy-murattal-hafs` | Mishari Rashid Al Afasy (Murattal – Hafs) |
| 23 | `mohamed-al-tablawi-murattal-hafs` | Mohamed Al Tablawi (Murattal – Hafs) |
| 24 | `muhammad-al-luhaidan` | Muhammad Al Luhaidan |
| 25 | `muhammad-siddiq-al-minshawi-kids` | Muhammad Siddiq Al Minshawi (Kids) |
| 26 | `muhammad-siddiq-al-minshawi-murattal-hafs` | Muhammad Siddiq Al Minshawi (Murattal – Hafs) |
| 27 | `nabil-ar-rifai` | Nabil Ar Rifai |
| 28 | `noreen-siddiq-duri-an-abu-amr` | Noreen Siddiq (Duri 'an Abu Amr) |
| 29 | `raad-mohammad-al-kurdi` | Raad Mohammad Al Kurdi |
| 30 | `saad-al-ghamdi-murattal-hafs` | Saad Al Ghamdi (Murattal – Hafs) |
| 31 | `saud-al-shuraim-murattal-hafs` | Saud Al Shuraim (Murattal – Hafs) |
| 32 | `surah-recitation-hady-toure` | Surah Recitation – Hady Toure |
| 33 | `tawfeeq-bin-saeed-as-sawaaigh` | Tawfeeq bin Saeed As Sawaaigh |
| 34 | `yasser-al-dosari-murattal-hafs` | Yasser Al Dosari (Murattal – Hafs) |

---

## 📜 Data License & Usage

**License:** `Apache-2.0`

This dataset is completely free and open-source for personal, educational, and commercial projects under the **Apache License 2.0**. You are encouraged to integrate this directly into mobile apps (Flutter, React Native, iOS, Android) and web platforms.

**Attribution:**
While not strictly required by the license, we appreciate attribution linking back to this dataset to support the ongoing development of open-source Islamic infrastructure.