Two-Way Feature Extraction for Speech Emotion Recognition Using Deep Learning