I was just reading about a trio of grad students who, using nothing but a 10-minute audio recording of a person typing, were able to recover 96% of what was typed. Very cool. It turns out you don’t even need to have a calibration pass to match the subtle variations in sound of each key.
Since there is one key for every letter, I’m guessing the students are doing a type of frequency analysis, a cipher-breaking technique dating back to the 9th century. It works like this: though at first you don’t know what key makes what sound, you do have a lot of sounds collected. For the English language, we know that certain letters are more likely to appear than others. This information is found in a “letter frequency table”, and you can make your own by counting letters in enough text. So see which letters seem to be occuring most often, and you can guess that those letters correspond to the top entries of the letter frequency table. Plug ’em in and see if it makes sense.
Also, in two- or three-letter words, we have a good chance of guessing that they’re words like THE, A, and IN. And once we have those letters matched, we have a good chance of uncovering other letters through context…if you’re good at “Wheel of Fortune”, you get the idea. Now imagine a computer doing this all for you. Scary!
Via Freedom To Tinker…quite some time ago :-)