Universal Adversarial Perturbations for Speech Recognition Systems
First Set (|V|∞ = 100): -42 dB distortion between original and perturbed audio
-
Example 1
-
Original
-
Original + Uniform Noise
-
Original + Universal Perturbation
-
Example 2
-
Original
-
Original + Uniform Noise
-
Original + Universal Perturbation
Second Set (|V|∞ = 150): -38 dB distortion between original and perturbed audio
-
Example 1
-
Original
-
Original + Uniform Noise
-
Original + Universal Perturbation
-
Example 2
-
Original
-
Original + Uniform Noise
-
Original + Universal Perturbation
Third Set (|V|∞ = 200): -36 dB distortion between original and perturbed audio
-
Example 1
-
Original
-
Original + Uniform Noise
-
Original + Universal Perturbation
-
Example 2
-
Original
-
Original + Uniform Noise
-
Original + Universal Perturbation
Fourth Set (|V|∞ = 300): -32 dB distortion between original and perturbed audio
-
Example 1
-
Original
-
Original + Uniform Noise
-
Original + Universal Perturbation
-
Example 2
-
Original
-
Original + Uniform Noise
-
Original + Universal Perturbation