Towards Event Extraction from Speech with Contextual Clues