Source allophony and speech synthesis