ELEGANCE: Efficient LLM Guidance for Audio-Visual Target Speech Extraction