Breaking Speaker Recognition with PaddingBack

arXiv:2308.04179v2 Announce Type: replace
Abstract: Machine Learning as a Service (MLaaS) has gained popularity due to advancements in Deep Neural Networks (DNNs). However, untrusted third-party platforms have raised concerns about AI security, particularly in backdoor attacks. Recent research has shown that speech backdoors can utilize transformations as triggers, similar to image backdoors. However, human ears can easily be aware of these transformations, leading to suspicion. In this paper, we propose PaddingBack, an inaudible backdoor attack that utilizes malicious operations to …

ai security arxiv attacks backdoor backdoor attacks backdoors breaking can cs.cr cs.sd eess.as eess.sp human image machine machine learning networks neural networks party platforms recognition research security service speaker speaker recognition speech third third-party untrusted