Over-the-air White-box Attack on the Wav2Vec Speech Recognition Neural Network
This work addresses a practical limitation in deploying adversarial attacks for speech recognition, but it is incremental as it builds on existing over-the-air attack methods.
The paper tackled the problem of making over-the-air adversarial attacks on Wav2Vec speech recognition systems less detectable by human hearing, and explored the trade-offs between detectability and attack effectiveness.
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.