EveGuard Introduces New Framework for Voice Privacy Protection
/ 3 min read
Quick take - EveGuard is a newly introduced software-driven defense framework designed to protect voice privacy from eavesdropping via vibrometry-based side channels, utilizing innovative techniques such as a perturbation generator model and Eve-GAN to effectively suppress audio reconstruction efforts while maintaining audio quality.
Fast Facts
-
Introduction of EveGuard: A new software-driven defense framework designed to protect against eavesdropping via vibrometry-based side channels using sensors like mmWave radars and accelerometers.
-
Innovative Approach: Unlike traditional hardware solutions, EveGuard utilizes a perturbation generator model (PGM) to create adversarial audio that suppresses eavesdropping while maintaining audio quality.
-
Eve-GAN Technology: Incorporates a domain translation task to enhance the PGM’s effectiveness, allowing for end-to-end training and minimizing data collection overhead through few-shot learning techniques.
-
High Protection Rate: Experimental results show EveGuard achieves over 97% protection against audio classifiers, effectively hindering audio reconstruction from eavesdropping.
-
Versatile Deployment: Designed for both offline and online use, EveGuard ensures low-latency performance for real-time applications and is feasible for deployment on cloud and on-device platforms.
EveGuard: A New Defense Framework Against Eavesdropping
EveGuard, a new software-driven defense framework, has been introduced to combat eavesdropping via vibrometry-based side channels. These side channels utilize sensors like mmWave radars, light sensors, and accelerometers, which have raised privacy concerns. Although these sensors are not designed for voice recording, they can capture vibrations from sound sources, leading to potential privacy breaches.
Innovative Approach to Voice Privacy
Traditional defenses against such eavesdropping often involve expensive hardware solutions with physical limitations. EveGuard offers a novel approach by protecting voice privacy without compromising human perception. Unlike traditional microphones that detect air pressure changes, EveGuard uses side-channel sensors to capture vibrations.
The framework employs a perturbation generator model (PGM) to produce adversarial audio, suppressing sensor-based eavesdropping while maintaining audio quality. A significant innovation in EveGuard is the introduction of Eve-GAN, a domain translation task that infers eavesdropped signals from audio. This innovation facilitates the end-to-end training of the PGM, enhancing its effectiveness. Few-shot learning techniques are incorporated to minimize data collection overhead during Eve-GAN’s training process.
Performance and Versatility
Experimental results show EveGuard achieves a protection rate exceeding 97% against audio classifiers, significantly hindering audio reconstruction efforts from eavesdropping. EveGuard’s performance has been validated across three adaptive attack mechanisms, demonstrating resilience under various conditions. User studies confirm the perceptual quality of the perturbed audio, with participants reporting high quality and difficulty in interpretation.
EveGuard addresses challenges such as ensuring effective perturbations and automating perturbation generation. Its architecture features a two-stage PGM using FIR convolution for low-frequency perturbations and inaudible low-frequency adversarial perturbations (LFAPs). EveGuard capitalizes on the transferability of adversarial examples to bolster robustness against unknown attack models.
Evaluation metrics for testing EveGuard’s effectiveness include Mel-Cepstral Distortion (MCD), Word Error Rate (WER), Digit Detection Rate (DDR), and Perceptual Evaluation of Speech Quality (PESQ). Results indicate EveGuard significantly increases MCD and WER, showing effective protection, while also reducing DDR, underscoring its efficacy against eavesdropping attempts.
EveGuard is designed for both offline and online scenarios, ensuring low-latency performance for real-time applications like VoIP communications. Its feasibility for deployment in cloud and on-device platforms makes it a versatile solution for voice privacy, particularly relevant as intelligent devices with sensors become more prevalent. EveGuard represents a significant advancement in voice privacy protection, introducing innovative techniques and robust methodologies to combat threats from side-channel eavesdropping technologies.
Original Source: Read the Full Article Here
