Automatic speech recognition using articulatory features from subject-independent acoustic-to-articulatory inversion.