Fig. 4From: Performance vs. hardware requirements in state-of-the-art automatic speech recognitionRWTH RETURNN (left) and PaddlePaddle DeepSpeech2 (right)Back to article page