A group of British researchers have developed an algorithm that can identify what someone is typing on a keyboard by the sound of the keystrokes alone.
Using standard microphones, this technology can detect keystrokes with an accuracy of 95%. When Zoom was used to train the algorithm, the accuracy dropped slightly to 93%, but is still alarmingly high, reports Bleeping Computer.
This technology could potentially be exploited by malicious actors to steal passwords and other sensitive information. Unlike traditional so-called “keyloggers”, which require the installation of malware on the victim’s computer, this method only needs access to the sound of keystrokes.
The researchers collected data by recording the sound of 36 keys being pressed 25 times each on a MacBook Pro. The audio recordings were then converted into spectrograms, which were used to train the image classification algorithm “CoAtNet”.
For those worried about this type of acoustic eavesdropping attack, however, there are safeguards. The researchers suggest that users can vary their typing style, use random passwords or play “white noise” while typing. The use of biometric authentication and password managers, which eliminate the need for manual input of sensitive information, can also be effective countermeasures.