metadata
license: bsd-3-clause
tags:
- speaker extraction
- speech separation
- gesture
- pose
- multimodal
- cocktail party problem
A modified version of the Youtube Gesture Dataset. Original data can be downloaded and processed here and here.
This dataset is used for Audio-visual speaker extraction conditioned on body gestures in the SEG paper, which the code can be found here.