Adversarial samples

KS

Parameters of Generative Adversarial Samples:

n_iters = 100, eps = 0.0005, sigma = 0.0005

KS	w2v2-w2v2	hubert-hubert	w2v2-hubert	hubert-w2v2	gau-w2v2	gau-hubert
acc	65.67	64.33	96.67	91.00	98.67	99.67
nsr	1.05	1.25	1.05	1.25	1.08	1.08
aer	100.77	95.00	42.55	71.17	40.59	19.84

HUBERT

Clean	IFGM	Gaussian

WAV2VEC

Clean	IFGM	Gaussian

IC

Parameters of Generative Adversarial Samples:

n_iters = 100, eps = 0.0005, sigma = 0.0004

IC	w2v2-w2v2	hubert-hubert	w2v2-hubert	hubert-w2v2	gau-w2v2	gau-hubert
acc	55.67	69.67	98.33	88.33	93.67	99.67
nsr	0.46	0.51	0.46	0.51	0.70	0.70
aer	104.54	91.76	47.80	70.43	47.45	29.30

HUBERT

Clean	IFGM	Gaussian

WAV2VEC

Clean	IFGM	Gaussian

SF

Parameters of Generative Adversarial Samples:

n_iters = 100, eps = 0.0007, sigma = 0.0005

SF	w2v2-w2v2	hubert-hubert	w2v2-hubert	hubert-w2v2	gau-w2v2	gau-hubert
F1	88.55	76.91	99.42	95.36	99.71	99.89
cer	20.19	36.54	1.62	8.70	0.71	0.25
nsr	0.65	0.70	0.65	0.70	0.70	0.70
aer	63.18	55.12	27.87	42.28	29.56	19.74

HUBERT

Clean	IFGM	Gaussian

WAV2VEC

Clean	IFGM	Gaussian

SID

Parameters of Generative Adversarial Samples:

n_iters = 100, eps = 0.0002, sigma = 0.00018

SID	w2v2-w2v2	hubert-hubert	w2v2-hubert	hubert-w2v2	gau-w2v2	gau-hubert
acc	79.33	76.00	89.33	89.67	93.33	95.00
nsr	0.43	0.45	0.43	0.45	0.53	0.53
aer	36.95	38.41	22.85	27.02	9.19	6.54

HUBERT

Clean	IFGM	Gaussian

WAV2VEC

Clean	IFGM	Gaussian

ASR & PR

Parameters of Generative Adversarial Samples:

n_iters = 100, eps = 0.0005, sigma = 0.00045

ASR	w2v2-w2v2	hubert-hubert	w2v2-hubert	hubert-w2v2	gau-w2v2	gau-hubert
wer	19.20	26.76	1.94	5.54	0.48	0.05
nsr	0.67	0.71	0.67	0.71	0.69	0.69
aer	81.15	70.84	41.27	58.09	30.40	20.05

PR	w2v2-w2v2	hubert-hubert	w2v2-hubert	hubert-w2v2	gau-w2v2	gau-hubert
per	28.32	18.67	2.21	5.09	1.11	0.42
nsr	0.67	0.7	0.67	0.71	0.69	0.69
aer	87.88	67.76	39.95	58.72	32.67	19.13

HUBERT(ASR)

Clean	IFGM	Gaussian

WAV2VEC(ASR)

Clean	IFGM	Gaussian

HUBERT(PR)

Clean	IFGM	Gaussian

WAV2VEC(PR)

Clean	IFGM	Gaussian

ER

Parameters of Generative Adversarial Samples:

n_iters = 100, eps = 0.0002, sigma = 0.00016

ER	w2v2-w2v2	hubert-hubert	w2v2-hubert	hubert-w2v2	gau-w2v2	gau-hubert
acc	79.33	78.33	91.00	87.33	95.67	99.00
nsr	0.67	0.70	0.67	0.70	0.79	0.79
aer	55.02	55.71	33.48	40.18	17.68	10.87

HUBERT

Clean	IFGM	Gaussian

WAV2VEC

Clean	IFGM	Gaussian

SD

Parameters of Generative Adversarial Samples:

n_iters = 100, eps = 0.001, sigma = 0.00075

SD	w2v2-w2v2	hubert-hubert	w2v2-hubert	hubert-w2v2	gau-w2v2	gau-hubert
acc	88.84	88.27	95.36	94.45	98.22	98.37
der	16.73	17.39	6.69	8.21	2.53	2.29
nsr	1.43	1.50	1.43	1.50	1.47	1.47
aer	54.87	54.66	34.05	38.01	8.85	6.50

HUBERT

Clean	IFGM	Gaussian

WAV2VEC

Clean	IFGM	Gaussian

ASV

Parameters of Generative Adversarial Samples:

n_iters = 100, eps = 0.0006, sigma = 0.0004

ASV	w2v2-w2v2	hubert-hubert	w2v2-hubert	hubert-w2v2	gau-w2v2	gau-hubert
acc	91.67	88.33	96.67	97.00	99.00	99.67
nsr	1.13	1.18	1.13	1.18	1.16	1.16
aer	59.52	58.29	41.93	45.48	17.32	14.33

HUBERT

Clean	IFGM	Gaussian

WAV2VEC

Clean	IFGM	Gaussian