The segmentation doesn't really work when only auto-generated captions are available, often chopping off a complete sentence right from the middle, making the whole thing on sense at all. It really needs more work as it is a pretty important feature for users like me who's primary focus is listening practice. Hope this will be fixed soon