EURASIP Journal on Audio, Speech, and Music Processing

Table 4 Evaluation performance of centralized training and PFL training on the speaker identification task

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

Groups	Centralized [14]				PFL (Ours)
	ACC	Precision	F1-score	AUC	ACC	Precision	F1-score	AUC
G1	0.9909	0.9912	0.9908	0.9996	0.9664	0.9680	0.9654	0.9986
G2	0.6691	0.7945	0.6674	0.9785	0.7909	0.8138	0.7845	0.9862
G3	0.9747	0.9759	0.9746	0.9987	0.9379	0.9412	0.9368	0.9977
G4	0.3633	0.5805	0.3634	0.9254	0.6802	0.7151	0.6764	0.9695
G5	0.9696	0.9717	0.9696	0.9982	0.9421	0.9458	0.9416	0.9977
G6	0.4946	0.7006	0.5178	0.9439	0.7337	0.7819	0.7289	0.9802
G7	0.5837	0.6658	0.5606	0.9527	0.5565	0.6323	0.5359	0.9564
G8	0.2136	0.4647	0.2165	0.8485	0.3424	0.4301	0.3139	0.9092
G9	0.4904	0.6035	0.4809	0.9266	0.5088	0.5672	0.4877	0.9460
G10	0.2364	0.4085	0.2309	0.8540	0.3821	0.4584	0.3685	0.9177
G11	0.4630	0.6267	0.4800	0.9123	0.5257	0.5823	0.5160	0.9469
G12	0.2396	0.4812	0.2440	0.8448	0.4101	0.4917	0.4055	0.9244
Mean(\(\mu\))	0.5574	0.6887	0.5580	0.9319	0.6481	0.6940	0.6384	0.9609

Best performances are highlighted

Back to article page