Paper says domain-invariant models recover keystrokes from keyboard audio
Original title: Domain-Invariant and Language-Model-Assisted Acoustic Side-Channel Attacks on Keyboards
The authors say inference still works on unseen keyboards, users, noise, and compressed channels when a language model corrects predictions.
A cs.CR paper studies acoustic side-channel attacks that recover typed text from keystroke sounds captured by a commodity microphone, without access to the victim device. It argues that earlier results were limited to matched laboratory setups. The authors say keystroke inference survives unseen keyboards, diverse users, noisy rooms, and compressed channels when the acoustic model is domain-invariant and a language model corrects noisy predictions.
Unconfirmed: The open-world success claim is the paper's own result and is not independently confirmed here.
- No access to the victim device is required, according to the paper.
- Prior studies used one keyboard, a fixed microphone, and a quiet room.
- The claimed fix is domain invariance plus language-model correction.
- Compressed transmission channels are included in the open-world setting.
From the original article
1 min read at arxiv.orgAcoustic side-channel attacks (ASCA) recover typed text from the sound of keystrokes, using only a commodity microphone and no access to the victim's device.
By Bikrant Bikram Pratap Maurya · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.