Speech recognition using non-acoustical features