Hao-Ping (Hank) Lee (Carnegie Mellon University), Wei-Lun Kao (National Taiwan University), Hung-Jui Wang (National Taiwan University), Ruei-Che Chang (University of Michigan), Yi-Hao Peng (Carnegie Mellon University), Fu-Yin Cherng (National Chung Cheng University), Shang-Tse Chen (National Taiwan University)
Audio CAPTCHA (Completely Automated Public Turing test to tell Computers and Humans Apart) is an accessible alternative to the traditional CAPTCHA for people with visual impairments. However, the literature has found that audio CAPTCHA suffers from both lower usability and security than its visual counterpart. In this paper, we propose AdvCAPTCHA, a novel audio CAPTCHA generated by using adversarial machine learning techniques. By conducting studies with people with and without visual impairments, we show that AdvCAPTCHA can outperform the status quo audio CAPTCHA in security but not usability. We demonstrate AdvCAPTCHA’s feasibility of providing detection of malicious attacks. We also present an evaluation metric, thresholding, to quantify the trade-off between usability and security for audio CAPTCHA design. Finally, we discuss approaches to the real-world adoption of AdvCAPTCHA.