EchoLLM - LLM-Augmented Acoustic Eavesdropping Attack on Bone Conduction Headphones with mmWave Radar

Explore a groundbreaking research presentation on EchoLLM, an innovative acoustic eavesdropping attack that targets bone conduction headphones using millimeter wave radar technology. Learn how researchers from Central South University and University of Massachusetts Lowell developed the first mmWave-based attack specifically designed to intercept semantic content from audio transmitted through bone conduction devices by exploiting mechanical vibrations as a side channel. Discover the technical methodology behind this context-aware and target-aware attack system that incorporates a novel multi-modal speech recognition model leveraging the victim's own speech patterns for enhanced automatic speech recognition accuracy under weak mmWave signals. Examine the signal enhancement techniques employed, including target identification and background reflection reduction, that enable the system to achieve remarkably low word error rates of 5.23% in realistic scenarios. Understand the serious privacy implications this research reveals about bone conduction headphones, which have gained popularity for their comfort and versatility in transmitting sound via skull bone vibrations rather than through eardrums, and how this attack demonstrates the vulnerability of what users may consider secure audio transmission methods.