Augmenting short-term cepstral features with long-term discriminative features for speaker verification of telephone data.