A method and apparatus for generating subunit speech patterns