Enriching Multimodal Sentiment Analysis through Textual Emotional Descriptions of Visual-Audio Content