Modeling Turn-Taking in Human-To-Human Spoken Dialogue Datasets Using Self-Supervised Features.