A comparative study of bottom-up and top-down approaches to speaker diarization