I looked into stenography for a client who had a problem with audio being leaked by the users of his service. Even though there is a lot less signal in audio, it was trivial to encode a 32-bit pattern throughout the audio track in such a way that it was not discernible to his well-trained ears, and yet it would survive being encoded as a 64kbps MP3. We also checked pushing it through an analog loop, since that was most likely how his tracks were being pirated.
The pattern would repeat throughout the song (I think once per minute) such that only a small sample was necessary to recover the pattern. It also had enough check bits that it was impossible to get the wrong code; either you'd get back the right code or nothing. This was all done with COTS software (a command-line executable) that could be licensed for a few hundred dollars.
My client ultimately decided not to add stenography. He believed that his users would be upset with the possibility that there was some quality loss due to the process, as they were paying a lot of money to have high-definition tracks. He thought the losses from the discovery that he was applying stenography would exceed the ongoing loss from piracy.