Over-the-air White-box Attack on the Wav2Vec Speech Recognition Neural Network

TL;DR

This paper investigates making over-the-air adversarial attacks on Wav2Vec speech recognition models less detectable by humans, while analyzing how these modifications affect attack success.

Contribution

It introduces methods to reduce human detectability of over-the-air attacks and examines their impact on attack effectiveness.

Findings

Reduced human detectability of adversarial audio

Trade-offs between attack stealth and success rate

Insights into attack robustness in real-world scenarios

Abstract

Automatic speech recognition systems based on neural networks are vulnerable to adversarial attacks that alter transcriptions in a malicious way. Recent works in this field have focused on making attacks work in over-the-air scenarios, however such attacks are typically detectable by human hearing, limiting their potential applications. In the present work we explore different approaches of making over-the-air attacks less detectable, as well as the impact these approaches have on the attacks' effectiveness.