Skip to main content

Table 3 Comparing performance of different training strategies of PFL

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

Evaluation groups

FL Classical [4]

PFL Type-A

PFL Type-B

 

Direct-discriminator

Discriminator-projection

Feature-projection

 

EER

EER

EER

G1

5.297

5.037

5.481

G2

15.089

14.100

25.242

G3

6.935

6.975

7.239

G4

19.961

18.986

20.485

G5

6.375

6.680

7.438

G6

17.355

15.796

19.703

G7

15.799

16.012

16.902

G8

24.619

25.096

31.103

G9

17.153

17.443

20.49

G10

23.805

23.653

25.927

G11

17.448

17.871

21.632

G12

22.806

22.993

26.048

mean(\(\mu\))

16.053

15.887

19.137

  1. Best performances are highlighted