A HMM-based speech synthesis system using a new glottal source and vocal-tract separation method