Inferring speaker size from speech formant data : A theoretical basis for vocal-tract length estimation