Which visual descriptors are suitable for multi-modal interaction and how to integrate them via real-time video data analysis into a corpus-based concatenative synthesis sound system.