It was explicitly trained on the keyboard used in MacBooks, which is fairly specific, but covers a pretty large user base.
In theory they could train it on other specific keyboards, but it remains to be seen what other factors could affect it
Sanctus@lemmy.world 1 year ago
Theres no way this thing is guessing keyclicks by sound on any keyboard. Maybe a specific one. Especially with custom keyboards taking off. My canonkeys 60% sounds nothing like my completely custom elvish keyboard. An AI in this day and age is not ready for that.
It was explicitly trained on the keyboard used in MacBooks, which is fairly specific, but covers a pretty large user base.
In theory they could train it on other specific keyboards, but it remains to be seen what other factors could affect it
Which, has a very specific sound with the scissor switches and aluminum casings. Its not exactly your average logitech keyboard in an office.
AbouBenAdhem@lemmy.world 1 year ago
Given that it’s AI-trained it may be hard to say, but my guess is that it’s based on timing more than the unique sound of each separate key. Like certain sequences of keys probably have a predictable time between each stroke, based on how long it takes the relevant finger to travel to the next key after the previous one.
Sanctus@lemmy.world 1 year ago
That one I could believe more. Since keyboards have such a wide array of sounds its ptobably not using the envelope to determine the key.
6daemonbag@lemmy.dbzer0.com 1 year ago
On top of that, we understand the frequency of letters used in languages. By knowing both of these and correlating with recurring patterns of sounds, I can very much believe this can be leveraged against even custom mechanical keyboards with random keys attached
hellishharlot@programming.dev 1 year ago
So switching to dvorak or colemak would possibly help significantly