Automatic Turn Segmentation in Spoken Conversations