Extracting deep bottleneck features for visual speech recognition | IEEE Conference Publication | IEEE Xplore