Summarizing multiple spoken documents: finding evidence from untranscribed audio.