Wouldn't a simple re-encode break any sort of stenography?
You can also embed data in the audio channel by alternating between two imperceptible echo kernels for 1/0[2]. This is extremely compression resistant.
1: http://ieeexplore.ieee.org/xpl/articleDetails.jsp?tp=&arnumb...
2: http://www.slidefinder.net/a/audio_steganography_echo_data_h...
Lossy compression: discard all sub-threshold information.
Thus, in the general case it's fundamentally impossible. You can make a scheme that will survive certain known encoders and bitrates, but you can't make one that will survive all lossy compression.