Does the resulting speech quality improvement make a sophisticated concatenation of time-domain synthesis units worthwhile?