In this paper, we consider the problem of resource-efficient architectures for audio-visual automatic speech recognition (AVSR). Specifically, we complement our earlier work that introduced efficient convolutional neural ...