Segmenting pop melodies: A model comparison approach