this is super cool. I did some digging about current research on avsr using video and lip point cloud movement a while back. found projects like https://github.com/amanvirparhar/chaplin but this seems more promising!