Fig. 2From: Three-stage training and orthogonality regularization for spoken language recognitionThe visualization results of language space trained with a frozen encoder strategy (a) and an unfrozen encoder (b) strategy, respectivelyBack to article page