Modified-VQ Features for Speech Emotion Recognition