Discriminating between voiced and unvoiced parts of speech signals