Mic-E-Mouse, Covert Eavesdropping Through Computer Mice
10 hours ago
- #Optical Sensors
- #Eavesdropping
- #Cybersecurity
- Computer mice with high-performance optical sensors can be exploited to eavesdrop on users by detecting acoustic vibrations.
- The attack, named Mic-E-Mouse, uses signal processing and machine learning to reconstruct intelligible speech from low-quality signals.
- Consumer-grade mice with high-fidelity sensors are affordable and increasingly common, expanding the potential attack surface.
- The attack can be delivered via open-source applications or video games, making data collection appear benign.
- Mic-E-Mouse overcomes challenges like heavy quantization and non-uniform sampling to achieve effective auditory surveillance.
- Human speech frequencies (200Hz to 2000Hz) are detectable by the pipeline, as shown in the 'Speech Banana' diagram.
- The attack is performant, with an SI-SNR increase of +19, 80% speaker-recognition accuracy, and 16.79% WER in tests.
- Anonymous repository and data access were provided for review purposes.