A speech analysis method based on a glottal source model