Articulatory movement features for short-duration text-dependent speaker verification