Fig. 2From: Towards end-to-end speech recognition with transfer learningTwo structures of DNN hidden layers. a Maxout-dropout hidden structure with a bottleneck layer. b Maxout-dropout hidden structure without bottleneck layerBack to article page