Clicky Keys
January 20, 2009 7:47 PM Subscribe
Is is possible to record the sound of someone typing and then decipher what keys are being pressed?
Probably not. Larger keys like SPACE and ENTER (and the tempo of somebody hitting backspace, or releasing SHIFT) have relatively distinctive sounds, but the rest are much of a muchness. Who are you hamfistedly attempting to spy on? ;-)
posted by turgid dahlia at 7:59 PM on January 20, 2009
posted by turgid dahlia at 7:59 PM on January 20, 2009
Not that I know of. It used to be very possible with phones with unique per-number beeps. But a few minutes spent with my keyboard listening carefully, the vast majority of the keys sound, at least to the unaided ear, more or less absolutely identical. The differences for periphery keys the spacebar, and so on probably would differ a little with each keyboard.
By the time you worked out a system to do it, I'd reckon a small in-line keylogger or a camera hidden to look at the monitor would be much much easier.
posted by LucretiusJones at 8:00 PM on January 20, 2009
By the time you worked out a system to do it, I'd reckon a small in-line keylogger or a camera hidden to look at the monitor would be much much easier.
posted by LucretiusJones at 8:00 PM on January 20, 2009
That's what I get for not previewing. Technology: always trumping common sense.
posted by LucretiusJones at 8:02 PM on January 20, 2009
posted by LucretiusJones at 8:02 PM on January 20, 2009
The problem with that method is that they need to have access to the keyboard beforehand so they can sample all the sounds the keys make.
Also their accuracy rate is not good until they at least three sets of data to correct errors. This attack is mostly theoretical. The attacker would need your keyboard for 20 or 30mins before they could snoop you. Its just easier to insert a hardware keylogger or software keylogger if you have physical access to the computer for that much time.
posted by damn dirty ape at 8:08 PM on January 20, 2009
Also their accuracy rate is not good until they at least three sets of data to correct errors. This attack is mostly theoretical. The attacker would need your keyboard for 20 or 30mins before they could snoop you. Its just easier to insert a hardware keylogger or software keylogger if you have physical access to the computer for that much time.
posted by damn dirty ape at 8:08 PM on January 20, 2009
The problem with that method is that they need to have access to the keyboard beforehand so they can sample all the sounds the keys make.
I think you have the wrong end of the stick. A previous study required pretraining, but the subsequent programme worked by recognising the individual sound of each key, and matching their frequency and order with a database of english in order to correctly assign the letter values. Also, the three sets of data are just feedthroughs of the output once it has been human-corrected. You only need to record one person typing one thing once to have a reasonable chance of figuring out what they typed.
posted by Sova at 8:39 PM on January 20, 2009
I think you have the wrong end of the stick. A previous study required pretraining, but the subsequent programme worked by recognising the individual sound of each key, and matching their frequency and order with a database of english in order to correctly assign the letter values. Also, the three sets of data are just feedthroughs of the output once it has been human-corrected. You only need to record one person typing one thing once to have a reasonable chance of figuring out what they typed.
posted by Sova at 8:39 PM on January 20, 2009
Somebody has a password checker that doesn't just look at the characters, it analyzes the strike-frequency of the entire passphrase. In other words, it measures the time it takes the user to type the password as well as interkey times. Since people have individual ways of typing their passwords, this can help identify someone who's successfully managed to *get* the password, but who's not the person who *owns* the password.
posted by zpousman at 8:22 AM on January 21, 2009 [1 favorite]
posted by zpousman at 8:22 AM on January 21, 2009 [1 favorite]
As people have commented, yes, it is possible, even without pre-training on that keyboard, to determine which keys someone hits given a recording of them typing a body of (unknown but) largely English text.
The original work (may require IEEE access) in the area relied on pre-training on the keyboard, but later work substitutes a recording of 10 minutes of unknown typing and gains a recognition rate of 96%.
posted by JiBB at 8:36 AM on January 21, 2009
The original work (may require IEEE access) in the area relied on pre-training on the keyboard, but later work substitutes a recording of 10 minutes of unknown typing and gains a recognition rate of 96%.
posted by JiBB at 8:36 AM on January 21, 2009
This thread is closed to new comments.
posted by Science! at 7:57 PM on January 20, 2009 [5 favorites]