Fig. 6From: Performance vs. hardware requirements in state-of-the-art automatic speech recognitionNvidia time convolutional networks: Jasper with dense residual blocks (left) and QuartzNet with time-channel separable residual blocks (right)Back to article page