This paper describes a fully automated framework to generate realistic head motion, eye gaze, and eyelid motion simultaneously based on live (or recorded) speech input. Its central idea is to learn ...