Fig. 8From: A simulation study on optimal scores for speaker recognitionBetween-class (left) and within-class (right) covariances of the x-vectors in CNCeleb, transformed by LDA trained on VoxCelebBack to article page