Perhaps if it relied on strength of hitting the keys, this would be harder, though some of this data might be retrievable from an accelerometre.
There's a chance they can make that a non-trivial problem, like recovering plaintext given a hash.
The problem is, someone out there isn't doing at least one of the above suggestions (or others that I've forgotten), or someone will intercept the original input while it's in flight (no one uses unencrypted connections on open public wifi, right?) or make use of some other arbitrary way that passwords can currently get stolen. And once they get that original digital representation of your typing style, you're done. That method will never be secure for you again for any service, unless you spend the time to change your typing style (and remember your new style should not be at all predictable to someone who knows your old style).
It's an equal certainty that there'll be implementations around for as long as humans remain that stupidly include un-hashed authentication data.
It might be possible to have the server randomly choose what text to test your style on. That would also solve the secure storage. But it's a very hard problem.
Say someone hacks the db where my particular typing style signature is stored. Now they can skip the typing and send the signal my keyboard would have sent and authenticate as me to whatever service was using that db.
You can push the problem into hardware (authenticate the keyboard as well) but that's just making the same mistake twice (once the keyboard gets hacked...).
Part of a good authentication system is being able to change the locks on the door when someone loses a key, which is why biometric data is particularly unsuited for the purpose.