Semi-fragile digital speech watermarking for online speaker recognition

Table 1 BERs for various fragile speech watermarking techniques under Stirmark® attacks

Attack	Semi-fragile DWPT-QIM (proposed)	AbS [25]	LSF [23, 24]	DT-CWT [22]	Genetic [26]
No attack	0	0.03	0.05	0	0.07
AddBrumm	0.45	0.50	0.56	0.12	0.53
AddSinus	0.43	0.54	0.54	0.13	0.41
AddNoise	0.39	0.52	0.55	0.25	0.48
Stat1 (statistical distortion)	0.54	0.73	0.67	0.11	0.58
Stat2 (statistical distortion)	0.51	0.41	0.34	0.23	0.64
Smooth1 (simple smoothing)	0.61	0.59	0.49	0.32	0.49
Smooth2 (simple smoothing)	0.55	0.53	0.61	0.10	0.42
Amplify (increases amplitude)	0.02	0.42	0.57	0.13	0.36
Invert (180° phase shift)	0.58	0.49	0.43	0.04	0.54
Exchange (swap samples)	0.63	0.54	0.65	0.03	0.23
CutSamples (7 samples per 1000)	0.41	0.51	0.71	0.32	0.45
LSBZero (resets LSBs)	0.37	0.44	0.51	0.02	0.65
ZeroCross (resets samples)	0.56	0.47	0.49	0.08	0.48
ZeroRemove (removes 0 samples)	0.5	0.45	0.52	0.17	0.44
Average	0.4367	0.4780	0.5127	0.1367	0.4513