Towards Data Distillation for End-to-end Spoken Conversational Question Answering