Elyadata/Ara-Best-RQ_dataset

---
license: cc-by-4.0
language:
- ar
multilinguality:
- monolingual
pretty_name: Ara-Best-RQ Dataset
size_categories:
- 1M<n<10M
---

# Ara-Best-RQ Dataset

## Dataset Summary

This dataset provides **metadata only** for a dialectal Arabic speech corpus constructed from publicly available YouTube videos.
It consists exclusively of **YouTube video identifiers** and **audio segment boundaries (start/end timestamps)** designed for **self-supervised speech representation learning**.

**No audio or video content is distributed as part of this dataset.**

---

## Dataset Statistics

- **Total spoken duration:** 5,639 h 04 min 27 s
- **Number of videos:** 41,826
- **Number of speech segments:** 3,860,419

---

## Supported Tasks

- Self-supervised speech representation learning
- Speech segmentation
- Arabic speech and dialect research

---

## Languages

- Arabic (`ar`)

---

## Dataset Structure

Each entry includes:
- A YouTube video identifier
- Segment start time (seconds)
- Segment end time (seconds)

Users are expected to retrieve the referenced media themselves and extract the corresponding audio segments.

---

## Licensing Information

### Metadata License

All metadata (video identifiers and segment boundaries) is released under:

**Creative Commons Attribution 4.0 International (CC BY 4.0)**

This license applies **only to the metadata** provided in this dataset.

### Referenced Media

The referenced audio/video content:
- Is **not distributed**
- Remains hosted on YouTube
- Is subject to the **original license selected by the content uploader**

Users are responsible for ensuring compliance with the original content licenses and YouTube’s Terms of Service.

---

## Citation

If you use this dataset, please cite the following paper:

> **Ara-Best-RQ: Multi Dialectal Arabic SSL**
> Haroun Elleuch, Ryan Whetten, Salima Mdhaffar, Yannick Estève, Fethi Bougares
> *Accepted at ICASSP 2026*

A BibTeX entry will be provided once the proceedings are published.

---

## Ethical Considerations

- This dataset references only publicly available content
- No personally identifying information is added or inferred
- No attempt is made to identify speakers

---

## Limitations

- Some referenced videos may become unavailable over time
- Audio quality and recording conditions vary
- Licensing information for individual videos is not exhaustively archived

---

## Contact

For questions, corrections, or takedown requests, please contact the dataset maintainer via the Hugging Face repository.