Extraction of phoneme-dependent individuality using HMM-based segmentation for text-independent speaker recognition