Modeling random uncertainties in voice production using a parametric approach