license: cc-by-4.0
task_categories:

• text-to-speech
  language:
• en
  size_categories:
• 10K<n<100K

UncovAI TTS Dataset

UncovAI TTS consists of 12,000 synthetic audio files generated from the DailyDialog dataset. It is designed for AI audio detection, forensics, and TTS research.

📂 Dataset Structure

The dataset is organized into three folders, with 4,000 audio files per model:

• Dia2-2B/: Generated using nari-labs/Dia2-2B
• Maya1/: Generated using maya-research/maya1
• MeloTTS-English/: Generated using myshell-ai/MeloTTS-English

🎯 Use Cases

• AI Audio Detection: Training classifiers to detect synthetic voices.
• Knowledge Distillation: Using high-quality synthetic data to train smaller student models.
• TTS Evaluation: Benchmarking different architectures on conversational text.

📜 Citation

If you use this dataset, please cite this repository, the DailyDialog dataset and the original model creators:

📄 Paper

Title: Audio Deepfake Detection in the Age of Advanced Text-to-Speech models
Authors: Robin Singh, Aditya Yogesh Nair, Fabio Palumbo, Florian Barbaro, Anna Dyka, Lohith Rachakonda
Paper: https://arxiv.org/abs/2601.20510

BibTeX

@dataset{uncovai2026uncovaitts,
  author       = {UncovAI},
  title        = {{UncovAI\_TTS}: Synthetic and real multilingual text-to-speech dataset},
  publisher    = {Hugging Face Datasets},
  year         = {2026},
  doi          = {10.57967/hf/7548},
  url          = {https://huggingface.co/datasets/UncovAI/UncovAI_TTS}
}

This project was provided with computing AI and storage resources by GENCI at IDRIS thanks to the grant 2025-AD011016076 on the supercomputer Jean Zay's V100 partition .