Going In Style: Audio Backdoors Through Stylistic Transformations

TL;DR

This paper introduces a novel audio backdoor attack method called JingleBack, which uses stylistic transformations like guitar effects to trigger malicious behavior with high success rates, highlighting new vulnerabilities in audio models.

Contribution

It formalizes stylistic triggers in audio backdoors and proposes JingleBack, a new attack method leveraging stylistic transformations for effective backdoor attacks.

Findings

Achieved a 96% attack success rate.

Formalized stylistic triggers in audio backdoor attacks.

Demonstrated effectiveness of guitar effects as triggers.

Abstract

This work explores stylistic triggers for backdoor attacks in the audio domain: dynamic transformations of malicious samples through guitar effects. We first formalize stylistic triggers - currently missing in the literature. Second, we explore how to develop stylistic triggers in the audio domain by proposing JingleBack. Our experiments confirm the effectiveness of the attack, achieving a 96% attack success rate. Our code is available in https://github.com/skoffas/going-in-style.