From Weak to Strong Sound Event Labels using Adaptive Change-Point Detection and Active Learning