Fig. 4From: Learning domain-heterogeneous speaker recognition systems with personalized continual federated learningComparing the error rate between the FL-classical and the PFL Type-ABack to article page