Physics‐in‐the‐Loop Evolutionary Adversarial Attack to Automatic Speech Recognition Systems under Hard‐Label Black‐Box Conditions

Key Points

• This research aims to explore methods for generating adversarial examples that can deceive automatic speech recognition systems in practical settings.
• Proposes a physics-in-the-loop approach for generating adversarial examples.
• Focuses on hard-label black-box conditions for testing.
• Assesses the impact of slight perturbations on prediction accuracy.
• Identifies vulnerabilities in automatic speech recognition systems under adversarial conditions.
• Demonstrates the effectiveness of the proposed attack method in achieving incorrect predictions.

Abstract