SamuriHL
UHD Guru
- Joined
- Jan 28, 2007
- Messages
- 15,032
- Likes
- 2,529
Just some speculation.
The watermark is embedded in the actual aduio itself.
The method used needs to be inaudible ( like tweeking a LS-bit every so often ). But that's not good enough, because it needs to be robust enough to be detectable in a microphone camcorder recording.
So they are not just bit-twiddling. That wouldn't survive the background noise on a camcorder recording, or even a re-encode.
I suspect they are actually creating a very low-rate data stream by introducing detectable acoustic signatures, below the noise-floor ( for them, the noise floor is the program audio. ) I think it likely that they are using correlation techniques. I suspect they are working in the frequency domain, adding a known spectrum of energy peaks at certain frequencies. The player will perform fourrier analysis on the audio stream, to give a spectrum. This will be correlated against the known spectrum, and where the correlellogram goes above some threshold, that will be a '1'. Since the signal level will be very low, the correlation window lengths will likely be large, and also averaged over a longish time ( perhaps tens of seconds ).
This gives a low data rate, which accounts for the times it takes to kick in.
To kill it, we'd need to know the correlation they are looking for. That's reverse-engineering. Once we can read their data stream by observing the correlation peaks come and go, we'd then need to apply carefull filtering to knock back the energy peaks which account for the correlation. You minimise the filtering to reduce the correlation peak to only below the detection threshold.
Then the trivial matter of re-encoding the audio
--
SC
Ummmmmm, hmmmmm. Yea, uh, about that. I think I'm going to stick with my MKV/ffdshow solution.