KFUPM-JRCAI/arabic-acoustic-poetry-baseline

Arabic Acoustic Poetry Dataset

Dataset Description

This dataset is a comprehensive collection of recited Arabic poetry for poem meter classification from acoustic sources. It contains 3,668 annotated audio files of recited Arabic poetry, where each sample represents a single Bait (poem verse) belonging to one of the 16 classical Arabic poetry meters. The recordings were collected in controlled laboratory conditions using professional-grade recording equipment.

Key Features

• Total Samples: 3,668 annotated audio files
• Duration: 9.277 hours (approximately 9 hours and 17 minutes)
• Average Sample Length: 8.985 seconds
• Storage Size: 897 MB
• Meters Covered: All 16 classical Arabic poetry meters
• Recording Quality: Laboratory-controlled conditions with professional equipment
• Language: Arabic (Classical Poetry)

Dataset Statistics

Metric	Value
Number of files	3,668
Total duration	9.277 hours
Average file length	8.985 seconds
Storage size	897 MB
Number of meters	16
Train samples	2,934
Validation samples	367
Test samples	367

Dataset Structure

HuggingFace Dataset Format

The dataset is organized into three splits:

DatasetDict({
    train: Dataset({
        features: ['Script', 'Bahr', 'Source', 'audio'],
        num_rows: 2934
    })
    validation: Dataset({
        features: ['Script', 'Bahr', 'Source', 'audio'],
        num_rows: 367
    })
    test: Dataset({
        features: ['Script', 'Bahr', 'Source', 'audio'],
        num_rows: 367
    })
})

Features Description

• Script: Text transcription of the recited poem verse (in Arabic)
• Bahr: The poem meter label (one of 16 classical Arabic meters)
• Source: Source information for the audio sample
• audio: Audio recording in WAV format

Splits

• Train: 2,934 samples (80%) - For training models
• Validation: 367 samples (10%) - For hyperparameter tuning
• Test: 367 samples (10%) - For final evaluation

Each audio sample consists of:

• Audio file: WAV format recording of a recited poem verse
• Meter label: One of the 16 classical Arabic poetry meters
• Transcription: Text transcription of the recited verse

The 16 Classical Arabic Poetry Meters

The dataset includes samples from all 16 classical meters defined in Arabic prosody ('Arud):

1. الطويل (Al-Taweel)
2. المديد (Al-Madeed)
3. البسيط (Al-Baseet)
4. الوافر (Al-Wafer)
5. الكامل (Al-Kamel)
6. الهزج (Al-Hazaj)
7. الرجز (Al-Rajaz)
8. الرمل (Al-Ramal)
9. السريع (Al-Saree')
10. المنسرح (Al-Munsareh)
11. الخفيف (Al-Khafeef)
12. المضارع (Al-Mudhare')
13. المقتضب (Al-Muqtadhab)
14. المجتث (Al-Mujtath)
15. المتقارب (Al-Mutaqareb)
16. المتدارك (Al-Mutadarek)

Citation

The dataset is described in detail in the paper:

"Poem Meter Classification of Recited Arabic Poetry: Integrating High-Resource Systems for a Low-Resource Task"

Available at: https://arxiv.org/pdf/2504.12172

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

@article{alshaibani2025poem,
  title={Poem Meter Classification of Recited Arabic Poetry: Integrating High-Resource Systems for a Low-Resource Task},
  author={Al-Shaibani, Maged S. and Alyafeai, Zaid and Ahmad, Irfan and Al-Zahrani, Abdul Kareem Saleh},
  journal={arXiv preprint arXiv:2504.12172},
  year={2025}
}

License and Usage

IMPORTANT DISCLAIMER

This dataset is published for academic and research purposes only.

Commercial use needs an explicit permission from the authors.

Commercial Licensing

For any commercial use of this dataset, you MUST contact the dataset owner to obtain proper licensing:

Dataset Owner: Dr. Abdul Kareem Saleh Al-Zahrani
Email: [email protected]
Faculty Page: https://faculty.kfupm.edu.sa/ias/akareem/index.htm

Unauthorized commercial use of this dataset is not allowed and may result in legal consequences.

Related Datasets

• Arabic Acoustic Poetry Benchmakr: https://huggingface.co/datasets/KFUPM-JRCAI/arabic-acoustic-poetry-benchmark/settings
• MetRec Dataset: Large-scale textual Arabic poetry dataset with 55,400 verses

Issues and Contributions

For issues, questions, or suggestions regarding this dataset, please contact [email protected].