Text-independent Speaker Identification Based on MAP Channel Compensation and Pitch-dependent Features