Dealing with Spoken Requests in a Multimodal Question Answering System