Learning to Segment Speech Using Multiple Cues: A Connectionist Model