Statistics for Frequency Domain Linear Prediction-Based Robust Text-Dependent Speaker Identification