TeethID: Teeth Gesture-Independent Earable Authentication Robust to Environmental Noise

Earbuds are increasingly used as human-centered computing platforms that store sensitive user data, creating a need for secure and unobtrusive authentication. Teeth-gesture-based authentication has emerged as a promising solution for implicit user verification, but existing approaches are often susceptible to environmental noise and variability in gesture execution, which can degrade authentication performance. This paper proposes a teeth-gesture-independent earbud authentication system (TeethID) that is robust to environmental noise. TeethID leverages the microphones embedded in commodity earbuds to capture teeth-gesture sounds from out-ear and in-ear microphones and constructs a user-specific out-ear-to-in-ear audio mapping function using predefined excitation signals, effectively suppressing environmental interference. In addition, TeethID incorporates an adversarial learning-based model to mitigate variations across different teeth gestures and inconsistencies across repeated occlusion instances. Extensive experiments involving 20 participants under diverse environmental noise levels demonstrate that TeethID achieves a balanced accuracy (BAC) of 97.11%, with a false acceptance rate (FAR) of 1.12% and a false rejection rate (FRR) of 4.66%. Security evaluations further show that TeethID is resilient to both replay and mimic attacks.