If you want a "real" watermark, not just some additional "meta data" that is easily removed, then that's
not trivial to do, indeed. You can't simply "mix" your watermark into the original signal. Instead, you have to manipulate certain "properties" of the audio signal, so that you will be able to distinguish "marked" from "unmarked" audio. And you need (at least)
two distinct "marked" states, if you want to embed an actual (binary)
message, instead of just being able to recognize whether a signal is "marked" or not. And, of course, this all needs to be done in a way that (A) doesn't impact audio quality too much, (B) survives the common attacks like MP3/AAC compression or analogue re-recording and (C) depends on a secret key.
There are commercial solutions available, but nothing Freeware or OpenSource that I'm aware of