High-Tech Gets a New Dimension

Researchers at Augusta University have introduced an acoustic attack that can identify key presses with 43% accuracy. This method, although less accurate than others, does not require controlled recording conditions or a special text input platform. The authors note that the method works even in noisy environments, providing enough reliable information for deciphering common input data through real-time analysis. The attack utilizes distinctive sound effects of different key presses and user text input patterns captured by specialized software for data collection. It is crucial to gather multiple text input samples from the victim to correlate specific key presses and words with sound waves. Researchers explore various input interception methods, including malicious software, websites, browser extensions, cross-site scripting, compromised applications, and USB keyboards. Input can be recorded using a hidden microphone in compromised devices like smartphones, laptops, or smart speakers. The captured dataset includes text input samples in various conditions, requiring multiple text input sessions crucial for the attack's success. However, these datasets do not need to be particularly large; they are used to train a statistical model that creates a detailed profile of individual text input patterns based on time intervals between key presses. A 5% deviation in the statistical model is crucial, as typing behavior changes slightly even when a person types the same word twice. The method predicts typed text by analyzing audio recordings of keyboard actions, with accuracy enhanced through filtering predictions using an English language dictionary. In addition to working in noisy conditions, the method is 43% effective in recording text input sessions on different keyboards, using a low-quality microphone, and any text input style. However, the method has limitations that reduce the attack's effectiveness. For instance, some people rarely use computers and have not developed a consistent text input pattern. The system also encounters issues with fast typing. Testing the method with 20 participants showed a wide range of success rates from 15% to 85%, indicating varying predictability among participants. According to researchers, the signal amplitude is less pronounced when using silent keyboards, reducing the model training effectiveness and key press detection speed.