Researchers at the University of Berkeley have developed a technique for keystroke detection based entirely upon audio sampling of the target. Reports indicate that after the algorithm has various filters applied in conjunction with spelling/grammar checkers, detection accuracy can reach upwards of 96%. Of course, new targets tend to have reduced accuracy in detection, but the algorithm learns with time the nature of the typist and can compare the sounds to a database of keyboard acoustic signatures. Such a technique offers surveillance (pdf) on subjects which can eliminate the need for blackbag operations to plant keyboard dongles or insert logging software on a computer. Such operations are already questioned by the legal community as invasive whereas this passive approach can be tied into existing audio monitoring via lasers. Combine these new innovations with older techniques like TEMPEST and you can passively recreate an interactive user session complete with keystrokes and screen captures using nothing more than access to line of sight with a pane of glass.