So here's how the code works, or rather is supposed to work (my code had an error that I only realized in retrospect due to code loss caused by
Visual Studio being a piece of loving stuff):
The lowest frequency is the signal frequency - this is on whenever a character is being transmitted. It's the repeating dash pattern highlighted below:

Then, the remaining frequencies encode a value. The lowest frequency encodes multiples of 25, the middle encodes multiples of 5, and the highest multiples of 1.
Certain patterns correspond to specific values, as illustrated below:

The values are here used to encode all ASCII values from 32 to 156 (which isn't an actual value under ASCII). One could drop three characters and ALMOST encode all of ASCII, but it might be better to have a Unicode-like system where if the low frequency has a value of 4 (values 100 - 124), then it corresponds to an extended character code. Due to the space character being value 0 in this system, it's quite visible using a visual frequency chart:

This may not actually line up with the previously posted sound files, because as I mentioned, the encoder I wrote previously had a bug.
EDIT: For reference, here's the same two messages as before, but re-encoded without the aforementioned bug.
Secret Message 1 &
Secret Message 2