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

Table 2

EER (%) of detection performance experiment with different methods and parameters.

Method	Feature	Parameter	Dev. set	Eval. set	Pooled
AESVDD-1	PCM	Dim = 64	30.31	1.42	11.39
	PCM	Dim = 128	29.32	1.63	11.91
	PCM	Dim = 256	28.45	2.71	12.81
	LFCC	Dim = 64	28.03	2.28	12.01
	LFCC	Dim = 128	27.24	5.09	12.32
	LFCC	Dim = 256	31.04	3.41	12.40
	CQT	Dim = 64	13.46	16.83	17.91
	CQT	Dim = 128	11.32	11.76	11.88
	CQT	Dim = 256	10.28	5.67	8.10
AESVDD-2	CQT	Dim = 64	21.47	3.89	10.88
	CQT	Dim = 128	23.12	1.24	10.12
	CQT	Dim = 256	13.5	5.97	9.20

Bold values indicate the optimal results.