Compensating the Speech Features via Discrete Cosine Transform for Robust Speech Recognition (基於離散餘弦轉換之語音特徵的強健性補償法)