---
license: cc-by-sa-4.0
tags:
- speaker extraction
- speech separation
- lip
- face
- multimodal
- cocktail party problem
---

A modified version of the VoxCeleb2 Dataset. Original data can be downloaded [here](https://www.robots.ox.ac.uk/~vgg/data/voxceleb/vox2.html). 

This dataset is used for Audio-visual speaker extraction conditioned on face recordings in [the reentry paper](https://ieeexplore.ieee.org/document/9721129), which the code can be found [here (ClearVoice repo)](https://github.com/modelscope/ClearVoice) or [here (Paper repo)](https://github.com/zexupan/reentry). 

# Usage

``` sh
cat orig* > orig.tar
tar -xvf orig.tar

cat audio_clean* > audio_clean.tar
tar -xvf audio_clean.tar
```