The study focuses on OverHear, a framework that infers keystrokes using both acoustic and accelerometer data from headphones. While the accelerometer data isn’t detailed enough for individual keystroke identification, it aids in clustering key presses by hand position. Simultaneously, the acoustic data is analyzed to extract Mel Frequency Cepstral Coefficients (MFCC), which helps distinguish between different keystrokes. These features feed into machine learning models for keystroke prediction, with the results further refined via dictionary-based word prediction methods. The results show a high level of prediction accuracy, highlighting both the effectiveness and limitations of this approach.
Publication date: 8 Nov 2023
Project Page: Not provided
Paper: https://arxiv.org/pdf/2311.02288
