This paper presents an HMM-driven hybrid speech synthesis approach in which unit selection concatenative synthesis is used to improve the quality of the statistical system using a Local Minimum Generation ...