Cocaine Noodles: Exploiting the Gap between Human and Machine Speech Recognition

Hands-free, voice-driven user input is gaining popularity, in part due to the increasing functionalities provided by intelligent digital assistances such as Siri, Cortana, and Google Now, and in part due to the proliferation of small devices that do not support more traditional, keyboardbased input. In this paper, we examine the gap in the mechanisms of speech recognition between human and machine. In particular, we ask the question, do the differences in how humans and machines understand spoken speech lead to exploitable vulnerabilities? We find, perhaps surprisingly, that these differences can be easily exploited by an adversary to produce sound which is intelligible as a command to a computer speech recognition system but is not easily understandable by humans. We discuss how a wide range of devices are vulnerable to such manipulation and describe how an attacker might use them to defraud victims or install malware, among other attacks.