Attacking Audio Event Detection Deep Learning Classifiers with White Noise

Abstract

We develop deep learning-based classifiers for Audio Event Detection (AED), attacking them next with some white noise disturbances. We show that an attacker can use such simple disturbances to potentially fully avoid detection by AED systems. Prior work has shown that attackers can mislead image classification tasks, however this work focuses on attacks against AED systems, by tampering the audio and not image. This work brings awareness to the designers and manufacturers of AED systems and devices, as these solutions are becoming more ubiquitous by the day.