Physics‐in‐the‐Loop Evolutionary Adversarial Attack to Automatic Speech Recognition Systems under Hard‐Label Black‐Box Conditions
Abstract
Adversarial examples are inputs with slight intentional perturbations that lead to incorrect predictions of deep neural networks; they also threaten automatic speech recognition systems. This study proposes a method to uncover such threats in practical environments, focusing on hard‐label black‐box settings. © 2026 Institute of Electrical Engineers of Japan and Wiley Periodicals LLC.
Fields of Study & Subjects
Adversarial Robustness in Machine Learning
Speech Recognition and Synthesis
Digital Media Forensic Detection
Publication details
- Journal / source
- IEEJ Transactions on Electrical and Electronic Engineering
- Volume
- 21
- Issue
- 10
- Pages
- 1578–1580
- Publication date
- 8 March 2026
- Language
- EN