I know next to nothing about steganography, beyond the general concept, so please excuse my ignorance if I'm way off base, but couldn't you design the encoding system such that you could encode the same amount of data within the time frame of each loop, with only some portion of the new data containing real data (audio recording)?
-edited because it didn't make sense before