Fig. 3From: Multiclass audio segmentation based on recurrent neural networks for broadcast domain dataAlternative BLSTM-based neural architectures including the proposed combination and pooling blockBack to article page