Robust Methods for Text-Dependent Speaker Verification