Adversarial Example Detection Techniques in Speech Recognition Systems: A review

Automatic Speech Recognition (ASR) is a crucial application of deep learning in today's world. However, ASR systems are vulnerable to attacks from malicious actors who can create adversarial examples to fool the system into producing incorrect outputs. Adversarial example detection techniques are being developed to mitigate the risk of such attacks. This review analyzes the effectiveness of various adversarial example detection techniques in the context of ASR systems. It begins by explaining the concept of adversarial examples and their potential impact on ASR systems as well as the most used evaluation metrics. Then, it provides an overview of the different types of adversarial attacks that can be launched against ASR systems. Upon reading this review, researchers are able to identify the existing weakness in ASR, and evade them in their future proposed schemes.