Black-Box Adversarial Attacks against Audio Forensics Models

Table 3

Adversarial attack success rate on decision-only black-box models (%).

Victim models	ResNet,	SEnet,	AFnet,	CNN-GRU,	Muti-models,	Muti-models,
ResNet		46	54	46	54	60
SEnet	40		38	42	53	62
AFnet	56	45		45	65	78
CNN-GRU	47	44	45		48	50

The bold values are the best results from our proposed method.