Conversations remotely detected from cell phone vibrations, researchers report
14 days ago
- #AI
- #privacy
- #surveillance
- Researchers at Penn State demonstrated 'wireless-tapping,' a method to remotely transcribe phone calls using radar measurements from up to 3 meters away.
- The technique captures vibrations from a phone's earpiece and uses AI-powered speech recognition to decode conversations, achieving ~60% accuracy for a 10,000-word vocabulary.
- The study builds on a 2022 project that identified predefined words with 83% accuracy using radar and voice recognition.
- Millimeter-wave radar sensors, similar to those in self-driving cars and 5G networks, were used in the experimental setup.
- The research aims to raise awareness about potential privacy risks posed by compact, radar-based surveillance devices.
- Open-source AI speech recognition models like 'Whisper' were adapted to interpret low-quality radar data.