The control of segmental duration in speech synthesis using linguistic properties