README.md

metadata

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.

This dataset is used for Audio-visual speaker extraction conditioned on face recordings in the reentry paper, which the code can be found here (ClearVoice repo) or here (Paper repo).

Usage

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

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