Urdu-ONYX-WAV

Urdu-ONYX-WAV is a high-quality Urdu Text-to-Speech (TTS) dataset consisting of audio recordings and corresponding transcripts. This dataset has been specifically prepared for training TTS models and conducting research in Urdu speech synthesis.

📊 Dataset Structure

This dataset is distributed across multiple parts due to size constraints:

• Main repository: Base dataset with initial samples
• part2: Additional 2.56 GB of audio data (6 Arrow files)
• part3:
• part N:

Combined Dataset Statistics

• Total Size: ~3+ GB (across all parts)
• Format: Apache Arrow (.arrow files)
• Audio Format: WAV, 22.05 kHz, 16-bit PCM
• Number of Samples: 100,000+ audio-transcript pairs

Data Fields

• id: Unique identifier for each sample
• transcript: Textual transcription of the audio (Urdu script)
• voice: Speaker identity or voice label
• text: Same as transcript (for TTS training convenience)
• timestamp: Recording timestamp (optional)
• audio: Audio data stored as bytes with metadata

🚀 Usage

Loading the Complete Dataset

To load all parts of the dataset:

from datasets import load_dataset, concatenate_datasets

# Load main dataset
main_dataset = load_dataset("humair025/Urdu-ONYX-WAV")

# Load additional parts
part2 = load_dataset("humair025/Urdu-ONYX-WAV", data_dir="part2")
# part3 = load_dataset("humair025/Urdu-ONYX-WAV", data_dir="part3")
# Add more parts as they become available

# Combine all parts
full_dataset = concatenate_datasets([
    main_dataset['train'],
    part2['train'],
    # part3['train'],
])

print(f"Total samples: {len(full_dataset)}")

Loading a Single Part

from datasets import load_dataset

# Load only part2
dataset = load_dataset("humair025/Urdu-ONYX-WAV", data_dir="part2")
print(dataset)

Accessing Audio Data

from datasets import load_dataset
import soundfile as sf

dataset = load_dataset("humair025/Urdu-ONYX-WAV")

# Access first sample
sample = dataset['train'][0]
print(f"Transcript: {sample['transcript']}")
print(f"Audio shape: {sample['audio']['array'].shape}")
print(f"Sample rate: {sample['audio']['sampling_rate']}")

# Save audio to file
sf.write('output.wav', sample['audio']['array'], sample['audio']['sampling_rate'])

📝 Citation Notice

⚠️ MANDATORY CITATION REQUIRED
You MUST cite this dataset in any publication, project, presentation, or derivative work — regardless of scope or scale. Proper attribution is a legal and ethical requirement under the modified Apache 2.0 license.

BibTeX Format

@misc{munir2025urduonyxwav,
  author       = {Humair Munir},
  title        = {Urdu-ONYX-WAV: A High-Quality Urdu Text-to-Speech Dataset},
  year         = {2025},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/datasets/humair025/Urdu-ONYX-WAV}},
  note         = {Multi-part dataset for Urdu speech synthesis}
}

APA Format

Munir, H. (2025). Urdu-ONYX-WAV: A High-Quality Urdu Text-to-Speech Dataset 
[Dataset]. Hugging Face. https://huggingface.co/datasets/humair025/Urdu-ONYX-WAV

📜 License

This dataset is released under a Modified Apache License 2.0 with mandatory attribution requirements.

Key Terms

1. ✅ Permitted Uses:

   • Academic research and publications
   • Commercial applications and products
   • Personal projects and experimentation
   • Model training and benchmarking
   • Modification and redistribution

2. ⚖️ Requirements:

   • Citation is mandatory for any use
   • Redistributions must include this README and license notice
   • Modified versions must be clearly marked as such
   • Attribution must be visible in papers, products, and documentation

3. 🚫 Restrictions:

   • No warranty or guarantee is provided
   • Author not liable for misuse or consequences
   • Users responsible for legal compliance in their jurisdiction

4. 📄 Full License:

   • Base license: Apache 2.0
   • Additional terms: Mandatory citation requirement

⚖️ Legal Notice

Content Disclaimer

• This dataset may contain synthetic audio, modified recordings, or human-recorded content
• The dataset is provided "AS-IS" without warranties of any kind
• Users assume all responsibility for:
  • Compliance with local laws and regulations
  • Proper use in accordance with ethical guidelines
  • Verification of content accuracy and quality

Redistribution Requirements

• All redistributions must include:
  • This complete README file
  • License notice and citation requirements
  • Acknowledgment of the original source

Liability Waiver

The dataset creator (Humair Munir) shall not be held liable for:

• Any damages arising from dataset use or misuse
• Errors, inaccuracies, or defects in the data
• Consequences of derivative works
• Legal issues arising from improper use

🎯 Recommended Applications

Primary Use Cases

• TTS Model Training: Tacotron2, VITS, FastSpeech2, Glow-TTS , StyleTTS2
• Speech Recognition: ASR model development and testing
• Voice Cloning: Speaker adaptation and voice conversion
• Linguistic Research: Urdu phonetics and prosody studies

Research Areas

• Low-resource language speech synthesis
• Multilingual TTS systems
• Speech quality assessment
• Urdu language processing

Benchmarking

• TTS model evaluation
• Speech synthesis quality comparison
• Cross-lingual transfer learning studies

🔧 Technical Specifications

Audio Properties

• Sample Rate: 22,050 Hz
• Bit Depth: 16-bit PCM
• Channels: Mono
• Format: WAV (uncompressed)

Data Format

• Storage: Apache Arrow format (.arrow files)
• Compression: Xet-optimized for efficient transfer
• Splitting: Multiple parts for manageable upload/download

Part Breakdown

Part	Size	Files	Status
Main	550 MB	2 Arrow files	✅ Available
part2	2.56 GB	6 Arrow files	✅ Available

and so on

📧 Contact & Contributions

Creator

Humair Munir

• Hugging Face: @humair025
• Dataset: Urdu-ONYX-WAV

Contributions

We welcome:

• Bug reports and corrections
• Quality improvement suggestions
• Additional Urdu speech data contributions
• Collaboration on Urdu TTS research

Please open an issue or discussion on the Hugging Face repository.

🙏 Acknowledgments

This dataset was created to support the development of high-quality Urdu speech synthesis systems and to contribute to low-resource language research.

---

📚 Related Resources

• Hugging Face Datasets Documentation
• TTS Training Guides
• Urdu NLP Resources

---

Last Updated: November 2025
Version: 1.0 (Multi-part release)
Status: Active development

---

If you use this dataset, please cite it using the BibTeX entry provided above. Thank you for contributing to responsible AI research!