Speaker Clustering Using Dominant Sets