Combined Visual Parameterization for Automatic Lip-Reading