Adapting Text-based Dialogue State Tracker for Spoken Dialogues