We introduce EmpathyNet, a dataset of video/image, audio and text cues, which can be easily compiled from existing video repositories.