(there's also a couple links at the end of this article discussing how she modified her radio to provide a digital interface for the data)
RDS at 57k she says? Stereo signal (or better, the L-R information) is at 38k (double band suppressed carrier, so the carrier would be at 38k). So 57 - 38 = 19k!
I wonder if this is a fairly common side channel attack? The most common in the domain of digital crypto is time (e.g. all password inputs must take equal time to check vs a given password, otherwise you leak info about the password) but distortion seems an obvious attack vector against analog crypto. Though I don't know if it's merely obvious in hindsight, as are most things.
The decryption is performed in a second step. The reason she had to do it this way is because most radios do not provide a convenient way to get at the subcarrier data.