Logan Blue

Automatic speech recognition and voice identification systems are being deployed in a wide array of applications, from providing control mechanisms to devices lacking traditional interfaces, to the automatic transcription of conversations and authentication of users. Many of these applications have significant security and privacy considerations. We develop attacks that force mistranscription and misidentification in state of the art systems, with minimal impact on human comprehension. Processing pipelines for modern systems are comprised of signal preprocessing and feature extraction steps, whose output is fed to a machine-learned model. Prior work has focused on the models, using white-box knowledge to tailor model-specific attacks. We focus on the pipeline stages before the models, which (unlike the models) are quite similar across systems. As such, our attacks are black-box and transferable, and demonstrably achieve mistranscription and misidentification rates as high as 100% by modifying only a few frames of audio. We perform a study via Amazon Mechanical Turk demonstrating that there is no statistically significant difference between human perception of regular and perturbed audio. Our findings suggest that models may learn aspects of speech that are generally not perceived by human subjects, but that are crucial for model accuracy. We also find that certain English language phonemes (in particular, vowels) are significantly more susceptible to our attack. We show that the attacks are effective when mounted over cellular networks, where signals are subject to degradation due to transcoding, jitter, and packet loss.

Voice interfaces are increasingly becoming integrated into a variety of Internet of Things (IoT) devices. Such systems can dramatically simplify interactions between users and devices with limited displays. Unfortunately, voice interfaces also create new opportunities for exploitation. Specifically, any sound-emitting device within range of the system implementing the voice interface (e.g., a smart television, an Internet-connected appliance, etc) can potentially cause these systems to perform operations against the desires of their owners (e.g., unlock doors, make unauthorized purchases, etc). We address this problem by developing a technique to recognize fundamental differences in audio created by humans and electronic speakers. We identify sub-bass over-excitation, or the presence of significant low frequency signals that are outside of the range of human voices but inherent to the design of modern speakers, as a strong differentiator between these two sources. After identifying this phenomenon, we demonstrate its use in preventing adversarial requests, replayed audio, and hidden commands with a 100%/1.72% TPR/FPR in quiet environments. In so doing, we demonstrate that commands injected via nearby audio devices can be effectively removed by voice interfaces.