Speechreading which infers spoken message from a visually detected articulated facial trend is a challenging task. In this paper, we propose an end-to-end ResNet (E2E-ResNet) model for synthesizing speech signals from the ...