The paper A Practical Deep Learning-Based Acoustic Side Channel Attack on Keyboards presents a practical implementation of a state-of-the-art deep learning model in order to classify laptop keystrokes. BleepingComputer’s write-up adds the setup: each of 36 keys on a MacBook Pro was pressed 25 times and recorded by an iPhone 13 mini 17 cm away, giving about 95% accuracy from the phone, 93% over Zoom and 91.7% over Skype. This already has a long research history with Don’t Skype & Type! Acoustic Eavesdropping in Voice-Over-IP and Researchers recover typed text using audio recording of keystrokes. One of the suggested mitigations is playing randomly generated fake keystrokes to the transmitted audio.
