Probabilistic Back-ends for Online Speaker Recognition and Clustering