---
license: mit
---
# Introduction  
MCL-MLAAD is the first multilingual benchmark for speech deepfake source tracing. It spans mono- and cross-lingual protocols, includes DSP and SSL baselines, studies language-specific fine-tuning for cross-lingual generalization, and tests robustness to unseen languages/speakers. See arXiv:2508.04143.

## Download the Dataset  

Install the `datasets` package:  
```bash
pip install datasets
```
Log in with your Hugging Face account:
```bash
huggingface-cli login
```
Load the dataset in Python:
```bash
from datasets import load_dataset

# Download from HF and cache
ds = load_dataset("xxuan-speech/MCL-MLAAD")

# Optionally: Save the dataset to your own directory
ds.save_to_disk("mcl-mlaad_local")
```

### Code: 
A reference implementation Code is available at:  https://github.com/xuanxixi/Multilingual-Source-Tracing

### Bibtex: 
```bash
@misc{xuan2025multilingualsourcetracingspeech,
      title={Multilingual Source Tracing of Speech Deepfakes: A First Benchmark}, 
      author={Xi Xuan and Yang Xiao and Rohan Kumar Das and Tomi Kinnunen},
      year={2025},
      eprint={2508.04143},
      archivePrefix={arXiv},
      primaryClass={eess.AS},
      url={https://arxiv.org/abs/2508.04143}
}
```