UC Irvine researchers demonstrate AI-enabled mouse sensors can capture speech
Researchers at UC Irvine developed a technique that uses AI and signal processing to convert acoustic vibrations captured by high-performance mouse sensors into speech, potentially enabling covert eavesdropping if a computer is compromised with data-collection malware.
What happened
Researchers at UC Irvine developed a technique called Mic-E-Mouse that uses AI to convert vibrations captured by high-performance mouse sensors into audible speech. High-DPI mice (20,000 DPI or above) can detect subtle acoustic vibrations transmitted through a desk surface when people speak nearby.
The attack requires the target computer to be compromised with malware capable of collecting high-frequency mouse data, which could be disguised as legitimate creative software or gaming applications that naturally require mouse sensor input. The raw sensor data can then be processed on another device using signal processing and machine learning.
Reported impact
Testing showed the technology achieved 42% to 61% speech recognition accuracy, with potential for improvement as AI technology advances. The research indicated that high-performance mouse sensors are now widely available at consumer price points ($100-$300) due to sensors like PAW3395/3399, expanding the potential attack surface.
Index analysis
The vulnerability exploits the high sensitivity and polling rates of gaming-focused mice, which can detect micro-vibrations as a side effect of their intended function. Signal processing techniques such as Wiener filtering combined with trained neural networks can extract speech signals from the acoustic vibration data.
Classification
Sources and evidence
Sources not yet published
This record was researched and written by the Index from original reporting. Its sources are still being verified by an editor, and they are listed here once that check is complete. Until then the record carries no reviewer.