In this paper, the speaker classification problem is addressed as an example of unknown-multiple signal source clustering. In this problem, the signal sequence corresponds to the LPC cepstrum, each category corresponds to a speaker, and the segmentation boundaries correspond to speaker transitions.