One paper accepted to WACV 2023! We introduce a new audio-video-language dataset of lecture videos and show that contrastive learning of narrations and video clip helps learn suitable representations to perform unsupervised lecture segmentation. ArXiv