Hugging Face's logo

Datasets:
NeoBoy
/
arabic-tts-wav-24k

Tasks:
Text-to-Speech
Modalities:
Audio
Text
Formats:
parquet
Languages:
Arabic
Size:
1K - 10K
Tags:
text2speech;
TTS;
Arabic
Libraries:
Datasets
Dask
License:
Dataset card Data Studio Files Community
1
arabic-tts-wav-24k / README.md
NeoBoy's picture
NeoBoy
Update dataset card with latest split/gender duration info
7d87655 verified
preview code
|
raw
history blame contribute delete
2.45 kB
metadata
license: mit
task_categories:
  - text-to-speech
language:
  - ar
tags:
  - text2speech;
  - TTS;
  - Arabic
pretty_name: STYLETTS2_ARABIC
size_categories:
  - 1K<n<10K

Arabic TTS WAV 24k Dataset

A high-quality, open-source dataset for Arabic Text-to-Speech (TTS) research, containing paired audio and text samples from both male and female speakers. All audio is provided in 24kHz WAV format, with rich metadata and phonetic transcriptions.

Dataset Summary

This dataset is designed for training and evaluating neural TTS systems in Modern Standard Arabic. It includes:

  • Audio: Clean, studio-quality WAV files at 24,000 Hz.
  • Text: Original Arabic text and phonetic (IPA) transcriptions.
  • Speakers: Both male and female voices, with gender labels.
  • Splits: Predefined train and test splits.
  • Metadata: Duration, sampling rate, file name, and source information for each sample.

Dataset Structure

Each sample contains:

  • text: The original Arabic sentence.
  • phonetic_text: IPA or phoneme transcription.
  • audio: Path to the 24kHz WAV file.
  • file: Unique file identifier.
  • gender: male or female.
  • sampling_rate: Always 24000.
  • duration: Duration of the audio in seconds.
  • source: Original dataset source.

The dataset is provided as a Hugging Face DatasetDict with "train" and "test" splits.

Split and Gender Statistics

Split Gender Files Duration (hrs)
train male 5963 7.00
train female 2593 5.29
test male 204 0.24
test female 81 0.17

Usage 

from datasets import load_dataset

dataset = load_dataset("NeoBoy/arabic-tts-wav-24k")
print(dataset)
print(dataset["train"][0])

Citation

If you use this dataset in your research, please cite the original sources:

And this dataset:

@misc{arabictts24k,
  title={Arabic TTS WAV 24k Dataset},
  author={Sharjeel Abid Butt},
  year={2025},
  howpublished={https://huggingface.co/datasets/NeoBoy/arabic-tts-wav-24k}
}

## License

This dataset is distributed for research purposes only. Please refer to the licenses of the original datasets for usage restrictions.

## Contact

For questions or contributions, please open an issue or contact me on github.com/NeoBoy .