Improving Speech-based Emotion Recognition with Contextual Utterance Analysis and LLMs