Learning domain-heterogeneous speaker recognition systems with personalized continual federated learning

EURASIP Journal on Audio, Speech, and Music Processing

Table 2 Evaluation performance of PFL and the baseline

Evaluation	Canonical model [9, 10]	Centralized training [14]	Separated training [26]	PFL (Ours)
Groups		(Assembling fine-tuning)	(Individual fine-tuning)	Type-A
	EER	EER	EER	EER
G1	2.85	9.664	13.712	4.978
G2	33.06	17.333	31.022	14.36
G3	4.81	10.627	14.174	6.864
G4	39.23	21.189	27.652	19.961
G5	5.51	10.351	17.977	6.274
G6	37.80	18.679	27.662	17.168
G7	16.25	18.86	24.728	15.412
G8	41.29	24.032	38.723	24.324
G9	20.61	18.975	30.294	16.554
G10	37.61	25.046	34.309	23.73
G11	24.45	19.419	33.983	16.779
G12	37.10	24.071	33.149	22.336
mean(\(\mu\))	25.0475	17.44	28.87	15.01