Dompteur: Taming Audio Adversarial Examples
This addresses a security threat in ASR systems used as hands-free interfaces, making attacks more detectable by humans, though it is an incremental approach focusing on perceptibility rather than elimination.
The paper tackles the problem of adversarial examples in Automatic Speech Recognition (ASR) systems by proposing a method that forces such examples to be audible to humans, using psychoacoustics to remove irrelevant information and train a model closer to human perception, resulting in adversarial examples that remain clearly perceivable with minimal computational overhead and preserved benign performance.
Adversarial examples seem to be inevitable. These specifically crafted inputs allow attackers to arbitrarily manipulate machine learning systems. Even worse, they often seem harmless to human observers. In our digital society, this poses a significant threat. For example, Automatic Speech Recognition (ASR) systems, which serve as hands-free interfaces to many kinds of systems, can be attacked with inputs incomprehensible for human listeners. The research community has unsuccessfully tried several approaches to tackle this problem. In this paper we propose a different perspective: We accept the presence of adversarial examples against ASR systems, but we require them to be perceivable by human listeners. By applying the principles of psychoacoustics, we can remove semantically irrelevant information from the ASR input and train a model that resembles human perception more closely. We implement our idea in a tool named DOMPTEUR and demonstrate that our augmented system, in contrast to an unmodified baseline, successfully focuses on perceptible ranges of the input signal. This change forces adversarial examples into the audible range, while using minimal computational overhead and preserving benign performance. To evaluate our approach, we construct an adaptive attacker that actively tries to avoid our augmentations and demonstrate that adversarial examples from this attacker remain clearly perceivable. Finally, we substantiate our claims by performing a hearing test with crowd-sourced human listeners.