Improving user interaction with voice synthesis