This research demonstrates how transformers and LLMs dramatically improve keyboard acoustic side-channel attacks, creating significant new security implications.

• CoAtNet model achieves state-of-the-art performance in keystroke identification
• Everyday microphones in common devices can now be weaponized more effectively
• Combines vision transformers with language models for enhanced accuracy
• Research identifies both attack vectors and potential mitigation strategies

As microphones become ubiquitous in homes and workplaces, this research highlights critical vulnerabilities in seemingly secure keyboard interactions, demanding new security approaches for both device manufacturers and end users.

Improving Acoustic Side-Channel Attacks on Keyboards Using Transformers and Large Language Models