README.md

---
license: cc-by-nc-4.0
task_categories:
- audio-classification
- automatic-speech-recognition
language:
- ar
size_categories:
- 100K<n<1M
tags:
- arabic
- arabic-speech
- ASR
- quran
arxiv: 2604.18932
pretty_name: 'Tadabur: A Large-Scale Quran Audio Dataset'
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*.parquet

- config_name: preview
  data_files:
  - split: train
    path: preview.parquet
---

<div align="center">

<img src="tadabur_logo.png" alt="Tadabur Logo" width="160" style="margin-bottom: 1rem;">

<h1>Tadabur: A Large-Scale Quran Audio Dataset</h1>

<p><em>The most comprehensive and richly annotated Qur'anic recitation corpus to date</em></p>

<p>
  <strong>Faisal Alherran</strong>
</p>

<p>
  <a href="https://arxiv.org/abs/2604.18932"><img src="https://img.shields.io/badge/📄_Paper-Tadabur-a27b5c?style=flat-square" alt="Paper"></a>
  &nbsp;
  <a href="https://github.com/fherran/tadabur"><img src="https://img.shields.io/badge/GitHub-fherran%2Ftadabur-1c1f1e?style=flat-square&logo=github" alt="GitHub"></a>
  &nbsp;
  <a href="https://huggingface.co/FaisaI"><img src="https://img.shields.io/badge/🎙_Models-FaisaI-374440?style=flat-square" alt="Models"></a>
  &nbsp;
  <a href="#license"><img src="https://img.shields.io/badge/License-CC_BY--NC_4.0-dfc49a?style=flat-square" alt="License"></a>
</p>

</div>

---

## ✦ Overview

**Tadabur** is a large-scale, high-diversity Qur'anic speech dataset designed to advance research in **Qur'anic Automatic Speech Recognition (ASR)**, **reciter modeling**, **tajwīd-aware speech processing**, and **prosodic analysis**. It is the most comprehensive publicly available collection of Qur'anic recitation audio to date.

---

## 📊 Dataset at a Glance

| Property | Value |
|:---|:---|
| 🕐 **Total Audio Duration** | 1400+ hours |
| 🎙️ **Distinct Reciters** | 600+ |
| 🔗 **Alignment** | Verse-level + word-level timestamps |
| ⚖️ **License** | CC BY-NC 4.0 |

---

## 🧭 Abstract

Despite growing interest in Quranic data research, existing Quran datasets remain limited in both scale and diversity. To address this gap, we present **Tadabur**, a large-scale Quran audio dataset comprising more than **1400 hours** of recitation audio from over **600 distinct reciters**, providing substantial variation in recitation styles, vocal characteristics, and recording conditions.

Tadabur includes complete coverage of all Qur'an, spanning styles such as *murattal* and *mujawwad*. Each file is accompanied by automatically derived **word-level temporal alignments** and structured metadata in a consistent JSON schema.

This diversity makes Tadabur a comprehensive and representative resource for Quranic speech research — enabling advances in ASR, tajwīd-aware modeling, reciter identification, and prosodic analysis.

---

## 📈 Comparison with Prior Datasets

| Dataset | Samples | Reciters | Transcription | Word-Level Alignment |
|:---|---:|---:|:---:|:---:|
| Quran Recitations (Kaggle) | 6,689 | 12 | ✗ | ✗ |
| Quran Speech-to-Text (SLR132) | 226,129 | 30 | ✓ | ✗ |
| Buraaq Quran Audio–Text | 187,080 | 30 | ✓ | ✗ |
| **Tadabur** *(Ours)* | **365,000+** | **600+** | **✓** | **✓** |

> Tadabur surpasses all prior publicly available Quranic datasets by a wide margin in scale, reciter diversity, and annotation richness.

---

## 🔄 Dataset Pipeline

A fully automated, multi-stage process transforms raw long-form recitations into clean, verse-level annotated audio files:

```
🌐 Collection  →  🤖 LLM Metadata  →  🎙️ WhisperX Align  →  ✂️ Boundary Detect  →  🧹 Curation
  Public repos      Surah & reciter      Word-level             Recitation-stop        ASR filtering &
  & archives        extraction via LLM   timestamps & AAM       segmenter              deduplication
```

---

## 🧩 Dataset Structure

Each example in the dataset consists of two components:

1. **Audio file** — a single Qur'anic verse (`.wav`)
2. **Metadata file (JSON)** — verse, reciter, and word alignment info

### JSON Schema

```json
{
  "reciter_id": 88,
  "surah_id": 3,
  "ayah_id": 82,
  "word_alignments": [
    {
      "word": "أَفَلَا",
      "start": 0.00,
      "end": 0.62
    },
    ...
  ],
  "text_ar_simple": "افلا يتدبرون القران ولو كان من عند غير الله لوجدوا فيه اختلافا كثيرا",
  "text_ar_uthmani": "أَفَلَا يَتَدَبَّرُونَ ٱلْقُرْءَانَ ۚ وَلَوْ كَانَ مِنْ عِندِ غَيْرِ ٱللَّهِ لَوَجَدُوا۟ فِيهِ ٱخْتِلَـٰفًا كَثِيرًا",
  "ayah_duration_s": 10.9,
  "audio_filename": "tadabur_spk0088_S3_A82_db1f8e71_000003.wav"
}
```

### File Naming Convention

```
tadabur_spk{reciter_id:04d}_S{surah_id}_A{ayah_id}_{hash}_{segment:06d}.wav
```

---

## 🎯 Supported Tasks
| Task | Description |
|:---|:---|
| 🗣️ **Automatic Speech Recognition (ASR)** | Verse-level transcription of Qur'anic audio |
| 👤 **Reciter Identification** | Speaker classification across 600 reciters |
| 📊 **Model Evaluation** | Benchmarking ASR models on Qur'anic speech using WER |

---

## 🤖 Fine-Tuned Whisper Models

Alongside the dataset, we release Whisper models fine-tuned on Tadabur for Qur'anic ASR. These models are domain-adapted to handle prolonged phoneme durations, tajwīd rules, melodic articulation, and the wide acoustic diversity unique to Qur'anic recitation.

| Model | Base | Status | Link |
|:---|:---|:---:|:---|
| **Tadabur-Whisper-Small** | Whisper Small | ✅ Available | [🤗 HuggingFace](https://huggingface.co/FaisaI/tadabur-Whisper-Small) |

---

## 📝 Notes

- Some reciters may appear **multiple times for the same surah and ayah**. These repetitions intentionally capture **natural variation** in recitation, including differences in pace, melody, pronunciation, and recording conditions.
- Recitation styles include *murattal* (measured/standard), *mujawwad* (melodic/ornate), and others.
- Audio was collected from public Qur'anic repositories and archives, then processed through the automated pipeline described above.

---

## ⚖️ License

This dataset is released under the **[Creative Commons Attribution–NonCommercial 4.0 (CC BY-NC 4.0)](https://creativecommons.org/licenses/by-nc/4.0/)** license.

**For research and educational purposes only. Attribution is required.**

---

## 🕌 Ethical & Cultural Considerations

The Qur'an holds deep religious and cultural significance for over a billion people worldwide. Tadabur is intended solely for **respectful and beneficial use**, including:

- ✅ Academic research
- ✅ Education and literacy tools
- ✅ Development of assistive technologies

🚫 Users must **not** employ this dataset for:
- Mockery or distortion of Qur'anic recitation
- Disrespectful or misleading audio synthesis
- Any application that may cause offense to Muslim communities

---

## 📚 Citation

If you use Tadabur in your research, please cite:

```bibtex
@misc{alherran2026tadabur,
  author        = {Alherran, Faisal},
  title         = {Tadabur: A Large-Scale Quran Audio Dataset},
  year          = {2026},
  eprint        = {2604.18932},
  archivePrefix = {arXiv},
  primaryClass  = {cs.SD},
  doi           = {10.48550/arXiv.2604.18932},
  url           = {https://arxiv.org/abs/2604.18932}
}
```

---

<div align="center">
  <sub>
    <strong>Tadabur</strong> · A Large-Scale Quran Audio Dataset ·
    <a href="https://github.com/fherran/tadabur">github.com/fherran/tadabur</a>
    <br>
    Released under CC BY-NC 4.0 for research and educational use.
    Users must engage with Qur'anic content respectfully.
  </sub>
</div>