F3SNet: A Four-Step Strategy for QIM Steganalysis of Compressed Speech Based on Hierarchical Attention Network

Table 3

Detection results under different embedding rates (sample length: and ).

Sample length (s)	Embedding rate (%)	ACC (%)	FNR (%)	FPR (%)
0.2	10	62.30	58.763	13.666
	20	74.71	37.451	13.417
	40	87.11	17.083	8.549
	60	94.38	7.892	3.271
	80	97.22	3.783	1.770
	100	98.88	1.130	1.116
1	10	71.19	33.845	23.582
	20	83.45	16.488	16.618
	40	94.63	4.762	5.962
	60	98.44	1.760	1.366
	80	99.51	0.869	0.098
	100	99.95	0.095	0