Modular BDPCA based visual feature representation for lip-reading | IEEE Conference Publication | IEEE Xplore