Fig. 5From: Three-stage training and orthogonality regularization for spoken language recognitionThe architecture of the end-to-end multi-task modelBack to article page