Continuously Predicting and Processing Barge-in During a Live Spoken Dialogue Task