The next frontier for computer-generated speech