How would you handle a situation where an AI agent fails to understand a candidate's response during a voice interview?