One-Class Fake Speech Detection Based on Improved Support Vector Data Description

Table 4

The result of cross-database experiment.

Method	EER (%)
	2019 dev.	In-the-wild	FoR
Res-TSSDNet + PCM	1.09	35.54	45.63
Res-TSSDNet + CQT	1.15	23.15	36.14
ResNet + spectrum	0.85	46.39	49.28
ResNet + MFCC	1.18	25.91	50.20
Proposed (AESVDD-1)	10.28	16.85	23.34
Proposed (AESVDD-2)	13.5	15.08	25.88

The left column includes results on ASVspoof 2019 development set, and the right columns are results on “In-the-wild” dataset and FoR dataset. Bold values indicate the optimal results.