Describe Where You Are: Improving Noise-Robustness for Speech Emotion Recognition with Text Description of the Environment.