Comparing Recognition Performance and Robustness of Multimodal Deep Learning Models for Multimodal Emotion Recognition