README.md

---
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](https://huggingface.co/datasets/roskoN/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](https://huggingface.co/nari-labs/Dia2-2B)
- `Maya1/`: Generated using [maya-research/maya1](https://huggingface.co/maya-research/maya1)
- `MeloTTS-English/`: Generated using [myshell-ai/MeloTTS-English](https://huggingface.co/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](https://huggingface.co/datasets/roskoN/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
```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 .