Designed and developed an efficient defense mechanism against voice command fingerprinting attacks on smart speakers

Personal Research

• Analyzed voice command network traffic from multiple smart speakers (Amazon Eco/Alexa, Google Home) to identify potential security vulnerabilities

• Developed a clustering-based defense strategy that uses traffic pattern recognition to effectively pad network flow with dummy packets

Achieved approximately 32% reduction in attack accuracy against baseline models

Currently conducting further experiments and preparing the full paper for submission to an international journal (IEEE TCAD)

Developed the first Deep Learning based Keyword Fingerprinting attack on the Tor network to infer user inputs, and concurrently engineered the BurstGuard defense mechanism to mitigate this novel threat.

Role : Data Acquisition / AI Researcher

• Engineered an automated traffic collection tool and configured the necessary environment for acquiring network traffic generated during keyword input on Tor.
  Successfully acquired a large-scale dataset, totaling 700,000 network traffic samples, for model training and evaluation.

• Designed and implemented the deep CNN architecture, consisting of five 1D Convolutional Blocks and two fully-connected layers for the core attack model, achieving 17% performance improvement over baseline model

Authored and published a full paper at ACM Asia Conference on Computer and Communications Security(AsiaCCS'25); August 2025

Analyzed network security of Amazon Alexa and Google Home; Inferring Voice Commands Using Machine Learning

Role : Data Acquisition / AI Researcher

• Led development of an automated Raspberry Pi + Alexa traffic capture pipeline. Orchestrated the collection of 500,000 network traces across 6 month for experiments

• Designed an window based feature and ensemble model architecture combining three feature-based models, achieving 38% performance improvement over baseline model

Authored and published a full paper at ACM Asia Conference on Computer and Communications Security (AsiaCCS '24); July 2024

Personal Research

• Automated network traffic data collection, yielding over 600,000 traces through the design and deployment of a new data pipeline

• Analyzed Zoom application-layer network traffic in depth and designed a CNN-based AI model to infer user activities during Zoom application usage

Awarded the Creative Researcher Scholarship Grant of Ewha Womans University