Hourglass-AVSR: Down-Up Sampling-Based Computational Efficiency Model for Audio-Visual Speech Recognition | IEEE Conference Publication | IEEE Xplore