Towards Incremental Speech Generation in Dialogue Systems