Self-supervised Cross-modal Pretraining for Speech Emotion Recognition and Sentiment Analysis F i n d i n g s - E M N L P