Research Article
Defense Mechanisms Against Audio Adversarial Attacks: Recent Advances and Future Directions
@INPROCEEDINGS{10.1007/978-3-031-28990-3_12, author={Routing Li and Meng Xue}, title={Defense Mechanisms Against Audio Adversarial Attacks: Recent Advances and Future Directions}, proceedings={Edge Computing and IoT: Systems, Management and Security. Third EAI International Conference, ICECI 2022, Virtual Event, December 13-14, 2022, Proceedings}, proceedings_a={ICECI}, year={2023}, month={3}, keywords={Adversarial defenses Deep learning Speech recognition Speaker recognition}, doi={10.1007/978-3-031-28990-3_12} }- Routing Li
Meng Xue
Year: 2023
Defense Mechanisms Against Audio Adversarial Attacks: Recent Advances and Future Directions
ICECI
Springer
DOI: 10.1007/978-3-031-28990-3_12
Abstract
With the popularity of speech and speaker recognition systems in recent years, voice interfaces are increasingly integrated into various Internet of Things (IoT) devices. However, studies have demonstrated that such systems are vulnerable to attacks using manipulated inputs. During the last few years, defense mechanisms have been studied and discussed from various aspects to protect voice systems from such attacks. Notwithstanding, there is lacking survey focus on the defense mechanism of audio adversarial examples. In this paper, we provide a comprehensive survey on state-of-the-art defense methods by illuminating their main concepts, reviewing the recent progress with a novel taxonomy, and discussing the future directions. It promises to bring awareness to the security problems in speech and speaker recognition systems and encourages people to propose more robust defenses against audio adversarial examples.
