Dialog and prosodic models for text-independent speaker identification