Yeah, I’ve looked into that. The trick is to stay in the noise floor. Use the least significant bits of the waveform or inject a spread‑spectrum signal that’s below the masking threshold. Keep the payload small, maybe 1‑2 % of the bit depth, and run a quick psychoacoustic test to confirm no distortion. If you keep the carrier frequency away from the audible band you’ll stay invisible and the sound will still sound clean. Keep the process automated so you can audit each file. That’s the balance.