US6760701B2 - Subword-Based Speaker Verification Using Multiple-Classifier Fusion, with Channel, Fusion, Model and Threshold Adaptation