Self-Learning Speaker Identification