Reproduce “Transforming Keystroke Noise to Text: Self-Supervised Acoustic Eavesdropping Attacks on Keyboards” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Atsunori Okada, Akira Ito, Rei Ueno, Yuichi Hayashi, Naofumi Homma
The authors present an audio-only method for reconstructing typed English from keyboard sounds by clustering clicks while retaining candidate characters. Character-level BERT and an LLM use linguistic context, while feedback refines the cluster-to-character mapping. In a controlled close-proximity test, it achieved a mean normalized Levenshtein score of 0.941 after 150 keystrokes, outperforming two baselines. This demonstrates evaluated English text, not arbitrary text.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.