Researchers have studied keystroke inference based on side channels, such as sound, electromagnetic wave, and timing. Since these attacks exploit characteristics of physical keyboards, they become ineffective on smartphones with soft keyboards.
Attacks using sensors on smartphone raises the awareness of privacy attacks on smartphone sensors. Besides the obvious privacy concern over the GPS sensor, researchers have shown attacks using the camera and microphone.
TouchLogger: Inferring Keystrokes On Touch Screen From Smartphone Motion is the first paper to show the privacy risks of motion sensors. Since typing on different locations on the screen causes different vibrations, motion data can be used to infer the keys being typed.
Both Android and iOS provide three accuracy levels based on event frequencies. For example, at the highest accuracy level, the average interval of device orientation events on an HTC Evo 4G phone is about 30ms, while that on a Motorola Droid phone is about 110ms. TouchLogger using motion sensor achieved an accuracy rate of over 70% on tests performed by researchers.