Visual-to-speech conversion based on maximum likelihood estimation | IEEE Conference Publication | IEEE Xplore