Interpretable Spectrum Transformation Attacks to Speaker Recognition Systems

Keyphrases

• Transferability 100%
• Spectral Transformation 100%
• Speaker Recognition System 100%
• White-box 100%
• Spectrum Transformation 100%
• Modified Discrete Cosine Transform 100%
• Black Box 66%
• Comparison Method 66%
• Attacker 66%
• Speaker Recognition 66%
• Surrogate Model 66%
• Class Activation Map 66%
• Time-frequency Domain 66%
• Gradient-based 33%
• Failed Attacks 33%
• Salient Region 33%
• Adversarial Attack 33%
• Saliency Map 33%
• Large Margin 33%
• Transfer Based 33%
• Ensemble Model 33%
• Adversarial Noise 33%
• Black Hole Attack 33%
• Audio Sample 33%
• Map Visualization 33%
• Ensemble of Surrogate Models 33%
• Transfer Attack 33%

Computer Science

• Recognition System 100%
• Speaker Recognition 100%
• Discrete Cosine Transform 100%
• Feature Map 66%
• Frequency Domain 66%
• Attackers 66%
• Successful Attack 33%
• Adversarial Machine Learning 33%
• Salient Region 33%
• Interpretability 33%
• Imperceptibility 33%
• Black-Box Attack 33%
• Comparison Result 33%
• Voice Input 33%