Researchers from British universities have developed a deep learning model that can steal data from keyboard keystrokes recorded using a microphone with 95% accuracy. Even when Zoom was used for training, the accuracy dropped only slightly to 93%, which is still alarmingly high. Acoustic attacks, such as this, pose a severe risk to data security as they could expose sensitive information, including passwords and messages. What makes this attack particularly dangerous is its simplicity and the availability of microphone-equipped devices. The researchers used spectrograms to train their image classifier, achieving high accuracy rates from smartphone recordings as well as Zoom and Skype calls. It is suggested that users alter typing styles or use randomized passwords as potential defense measures.
https://www.bleepingcomputer.com/news/security/new-acoustic-attack-steals-data-from-keystrokes-with-95-percent-accuracy/