EchoLLM: LLM-Augmented Acoustic Eavesdropping Attack on Bone Conduction...
summary
EchoLLM reconstructs bone‑conduction audio by combining a commercial mmWave radar + wireless mic with signal processing (range-beam selection, phase-to-displacement, circle‑fit reflection removal, LSTM spike interpolation, motion calibration) and a context‑aware multimodal ASR that uses the victim’s own speech as context. The prototype costs about $930 hardware cost, is tested across distances, angles, motions and four environments, achieves as low as 5.23% WER, and shows robustness across three headphone types.
AI-written summary · about
chapters (9)
- 0:00EchoLLM attack overview
- 2:47Millimeter wave radar use
- 4:57Threat model and goals
- 6:21System overview
- 7:46Vibration signal extraction
- 9:11Speech enhancement techniques
- 10:57Multimodal ASR framework
- 13:09Experimental setup and results
- 15:23Conclusion and future work
related talks
06mUsing Sonar for Liveness Detection to Protect Smart Speakers against Remote Attackers
12mSkullConduct: Biometric User Identification on Eyewear Computers Using Bone Conduction Through ...
06mEarDynamic: An Ear Canal Deformation Based Continuous User Authentication Using In-Ear Wearables
06mVoice In Ear: Spoofing-Resistant and Passphrase-Independent Body Sound Authentication
24mInaudible Voice Commands: The Long-Range Attack and Defense
45mAnalogy Cyber Security—From 0101 to Mixed Signals
35mThe Little Seal Bug: Optical Sound Recovery from Lightweight Reflective Objects
12mEarFisher: Detecting Wireless Eavesdroppers by Stimulating and Sensing Memory EMR