We present a novel resource-efficient end-to-end architecture for lipreading that achieves state-of-the-art results on a popular and challenging benchmark. In particular, we make the following contributions: First, inspired ...