Although it's claimed that this survives arbitrary transcoding, I'm really not convinced that's mathematically possible. I think it's much more likely that they've just chosen to embed information at a very low bit rate.
Having never heard of this before, I did the natural hackernews reader thing -- search for a bypass -- and came across this interesting forum discussion [1] from ~2014, in which a large number of people state that a reverb effect gets rid of it. The original author writes:
> get an program named Audacity open the converted audio file in freemake
> and open the file in Audacity choose File>Open then Edit>Select>All
> then go to Effect>GVerb
> make sure to have the following configuration:
> roomsize (m):1.0
> reverb time (s): 0.1
> Damping: 0.0
> Input Bandwidth: 0.20
> Dry Signal (db): -7.0
> Early reflection level (db): 0.0
> Tail Level: -17.5
Other methods of bypass include using a player that doesn't check for the watermark, patching out the checking code in a firmware image, and simply swapping the HD-DVD / Blueray audio with one from a DVD.
Some further useful information -- and a statement that they've put the details in their patent!
> The Cinavia detection is sensitive to pitches and time sequence of features. The specific feature that detector looks for, is already clearly stated in Verance patent US5940135:
> [2] www.google.com/patents/US5940135
> If you study the patent document you will know the feature is delayed correlation. They also use hopping to change the delay of the correlation within the pattern of the same watermark. The actual delay, and the hopping pattern between delays, is their secret and security. That information I cannot disclose. Nor is it needed to defeat Cinavia.
> The fact of the matter is Cinavia added an artificial signal to rapidly change the delayed correlation in short time internals. Analysis the audio and you can see it causes an un-natural ripples in the frequency space. If you can see those un-natural ripples, you can smooth it out and remove it. Just remember, in real world, different frequency components of a sound does not change that rapidly. They generally decay over a fraction of a second, and human ear can not catch it if a frequency shows up and then rapidly goes away. So the way to defeat Cinavia is do not let any frequency component go in and out so rapidly. If it shows up, let it stay for a little while longer. Stretch it out a bit, before letting it decay out.
> That will defeat Cinavia for sure. Make everything vary more slowly and smooth out any ripples. It also beautifies the sound quality.
> When people sing, the pace of their singing is much slower than normal speech, right? A sentence that takes 3 second to speak, a singer will spend half a minute to sing the same sentence out. The slow varying is what makes music beautiful, and it is also what can kill Civania!
Having never heard of this before, I did the natural hackernews reader thing -- search for a bypass -- and came across this interesting forum discussion [1] from ~2014, in which a large number of people state that a reverb effect gets rid of it. The original author writes:
> get an program named Audacity open the converted audio file in freemake > and open the file in Audacity choose File>Open then Edit>Select>All > then go to Effect>GVerb > make sure to have the following configuration:
> roomsize (m):1.0 > reverb time (s): 0.1 > Damping: 0.0 > Input Bandwidth: 0.20 > Dry Signal (db): -7.0 > Early reflection level (db): 0.0 > Tail Level: -17.5
Other methods of bypass include using a player that doesn't check for the watermark, patching out the checking code in a firmware image, and simply swapping the HD-DVD / Blueray audio with one from a DVD.
[1] https://web.archive.org/web/20141208081801/http://club.myce....
---- Edit ----
Some further useful information -- and a statement that they've put the details in their patent!
> The Cinavia detection is sensitive to pitches and time sequence of features. The specific feature that detector looks for, is already clearly stated in Verance patent US5940135:
> [2] www.google.com/patents/US5940135
> If you study the patent document you will know the feature is delayed correlation. They also use hopping to change the delay of the correlation within the pattern of the same watermark. The actual delay, and the hopping pattern between delays, is their secret and security. That information I cannot disclose. Nor is it needed to defeat Cinavia.
> The fact of the matter is Cinavia added an artificial signal to rapidly change the delayed correlation in short time internals. Analysis the audio and you can see it causes an un-natural ripples in the frequency space. If you can see those un-natural ripples, you can smooth it out and remove it. Just remember, in real world, different frequency components of a sound does not change that rapidly. They generally decay over a fraction of a second, and human ear can not catch it if a frequency shows up and then rapidly goes away. So the way to defeat Cinavia is do not let any frequency component go in and out so rapidly. If it shows up, let it stay for a little while longer. Stretch it out a bit, before letting it decay out.
> That will defeat Cinavia for sure. Make everything vary more slowly and smooth out any ripples. It also beautifies the sound quality.
> When people sing, the pace of their singing is much slower than normal speech, right? A sentence that takes 3 second to speak, a singer will spend half a minute to sing the same sentence out. The slow varying is what makes music beautiful, and it is also what can kill Civania!
[2] http://www.google.com/patents/US5940135