Watermarking: what do we know?

[andreasmaaan]

I'm no expert on this topic, which is why the title of this thread is a question, but it seems clear to me that a significant proportion of music distributed via streaming services is now watermarked by the major labels. Some examples of watermarked audio that have been posted in various places online seem to show that this practice does (or at least can) create quite audible distortion. The examples I've heard seemed to create more objectionable distortion than e.g. that caused by higher bitrate lossy compression, for example.

So I thought it might be worthwhile to try to develop a better understanding of what's going on here.

My questions are:

1. Which labels are doing this and how much of their catalogues is subject to it?
2. What different techniques are being used, and what are the technical and audible differences between these?
3. Is watermarking also being employed in respect of non-streaming releases?

Any solid information would be greatly appreciated.

 

[amirm]

My company (microsoft) participated in the first round of call for proposals for watermarks for music. For the longest time, content owners keep thinking this is a way of copy protection. It is not. All you can do is trace the content to the one with or without watermark. We (tech companies) adamantly fought the notion of using it for "copy control" and for the most part won. Proposals for example to only play music that has the official mark in them were thrown back at content owners and not implemented.

Back to watermark, it has multiple vectors:

1. How long of a mark needs to be inserted. 8 bits is easier than 64 bits. The more bits needed to be inserted, the more bits in music will need to change.

2. What is the shortest duration of music where the mark can be detected. Content owners want the shortest duration but from fidelity point of view, we want to hide the bits where they are least audible. If it takes 30 seconds, so be it. But this lengthens the forensic work to find the mark so there is always pressure to insert the mark/detect it in fewest seconds (e.g. 10).

3. How much CPU it takes to detect it. This is so that cheap embedded devices can find the mark and act on it. As is typically the case, simpler algorithm is less sophisticated in its ability to hide the mark.

4. Patent liability (i.e. risk of claims).

5. Cost of using the mark both in content and devices.

6. Resilience to damage due to data errors or transformation (e.g. encoding into MP3). For this reason, the mark data is hugely replicated. So what starts at 64 bits to be embedded, may turn out to be 64,000 bits you need to embed in the music.

All of these factors come into play before a label chooses a mark to embed. When at Microsoft, we proposed a sophisticated scheme that we thought did well in all of this but required a lot of CPU cycles. We lost out to a much simpler scheme that we thought was not sufficient at all.

I don't know what watermark technology is used these days. But the goal is always to reduce its audibility. To put it in context, it would be a lot easier to hear MP3 compression than a watermark. So I don't think most people will be bothered by it.

 

[andreasmaaan]

Thanks @amirm

Do you know whether watermarking is also applied to non-streaming releases?

And I’m not sure I fully understand the purpose of watermarking from a business perspective. Under which circumstances is it necessary to trace back the source of a recording from a business point of view?

Finally, on what basis do you believe that watermarking is always less audible than lossy compression?

I’ve read a couple of the early studies that are available online which discussed the audibility of particular watermarking techniques, but it wasn’t clear from those just how (in)audible these techniques were. The goal of lossy compression is also ofc to be as inaudible as possible, and much of the same psychoacoustic knowledge is applied in both cases. How can one be certain that it’s applied more successfully in the case of watermarking?

 

[amirm]

Do you know whether watermarking is also applied to non-streaming releases?

I don't have direct knowledge since I am not in the industry anymore.

 

[amirm]

And I’m not sure I fully understand the purpose of watermarking from a business perspective. Under which circumstances is it necessary to trace back the source of a recording from a business point of view?

Well, the one content owners wish is that every player in the world checks for the mark and then decides to play or not play. For example, they could set a mark in the streaming content that would say, "only play from streaming source." So if you stole the content from a streaming site and then distributed it to the world, it would have no value since nothing would play it. This is what we call "copy control" and industry turned down.

Another usage is to trace leakage. Let's say they put a different mark in the content they give to Amazon and Apple. They then scan the internet looking for those marks. To the extent they find 90% of the content has leaked from Amazon instead of Apple, they could go and complain to make their streaming service more secure against piracy.

Now, these were the old plans and dreams. The doors of copy protected were busted open when they conceded to remove DRM for content for Apple/iTunes. Since then, marks have no value in this domain.

There are other uses. For example, studios mark the audio track in movies they distribute to movie theaters. Blu-ray players (I think) are supposed to check that and not play content without that mark. Or play content that doesn't have the mark it is supposed to.

 

[maverickronin]

And I’m not sure I fully understand the purpose of watermarking from a business perspective. Under which circumstances is it necessary to trace back the source of a recording from a business point of view?

I think it's mostly a security blanket for suits with little technical understanding.

There isn't really a good use for marking most media in general distribution and I find it completely baffling that anyone would bother to watermark streaming media when unmarked media is available on CD or for paid download. Even if there weren't unmarked variants it's pretty useless for tracing the exact origin of anything in public distribution since most things are ripped a small number of times by pirate groups and then freely distributed. One group ripped it from Apple and another Amazon? And now someone cares why one version has 90% pirate market share? Why again? Also this assumes that pirate groups will bother redistributing lossy DRM free downloads, or even originally lossless but then watermarked streams as long as CDs exist.

The player locks that Amir mentions are the copyright cartel's wet dream, but have a snowball's chance in hell of ever actually happening.

The only thing it's really useful for is tracing leaks of limited release materials, like pre-release movie screeners. Then you can put a different mark on every copy and blackball whichever critic leaked (or got hacked) and let Avengers vs. Dark Justice League XVII get out a week early.

 

[hvbias]

Check out these two links:
https://www.mattmontag.com/music/universals-audible-watermark
http://mattmontag.com/audio-listening-test/

I have a hard time hearing it in some rock music, but I can hear it in classical music, particularly piano and string sections.

 

[andreasmaaan]

Check out these two links:
https://www.mattmontag.com/music/universals-audible-watermark
http://mattmontag.com/audio-listening-test/

I have a hard time hearing it in some rock music, but I can hear it in classical music, particularly piano and string sections.

Interesting, thanks. I got 12/16, p = 0.038 after launching more or less straight into it. That included 100% on the classical pieces. It's a bit frustrating that you can't see other people's results. Great test though.

Do you have any idea how representative´those particular samples are of watermarking techniques generally?

 

[hvbias]

Interesting, thanks. I got 12/16, p = 0.038 after launching more or less straight into it. That included 100% on the classical pieces. It's a bit frustrating that you can't see other people's results. Great test though.

Do you have any idea how representative´those particular samples are of watermarking techniques generally?

For the classical it is fairly similar to what I hear on Universal Music owned subsidary label (ie Deutsche Grammophon) streams when I had Tidal. I have also heard it on HDTracks downloads which is truly egregious.

Honestly not sure how people claiming to be audiophiles can not even mention it or say they can't hear it but are the same ones that have hundreds or thousands in cables. Or in a few instances say a hi-res download (with audible water marking to me) sounds better than the same mastering on CD with no watermark, because it's... hi-res.

You did better than me on that test, I am fairly sure I didn't get as high as 12

 

[andreasmaaan]

For the classical it is fairly similar to what I hear on Universal Music owned subsidary label (ie Deutsche Grammophon) streams when I had Tidal. I have also heard it on HDTracks downloads which is truly egregious.

Honestly not sure how people claiming to be audiophiles can not even mention it or say they can't hear it but are the same ones that have hundreds or thousands in cables. Or in a few instances say a hi-res download (with audible water marking to me) sounds better than the same mastering on CD with no watermark, because it's... hi-res.

You did better than me on that test, I am fairly sure I didn't get as high as 12

Ha, very wide margin of uncertainty with only 16 trials. Still, I scored better and felt more confident doing this test than I had felt while doing mp3 vs lossless blind tests in the past, which I think says something. I haven’t done those tests on my current (modest) headphone setup however, so can’t reallt compare the two.

Very disappointing to hear you found watermarking on an HD Tracks download

 

[maverickronin]

I have also heard it on HDTracks downloads which is truly egregious

Well I guess I'm never buying anything there again...

 

[hvbias]

Ha, very wide margin of uncertainty with only 16 trials. Still, I scored better and felt more confident doing this test than I had felt while doing mp3 vs lossless blind tests in the past, which I think says something. I haven’t done those tests on my current (modest) headphone setup however, so can’t reallt compare the two.

Very disappointing to hear you found watermarking on an HD Tracks download

I've always done rubbish when I've tried MP3 320 vs lossless. Way back in the day I fared better back when LAME was still calling the presets -aps or -ape (today's equivalent of -V2 and -V0 respectfully) I was able to pass a statistically significant test with -aps vs FLAC using some electronic music that someone hosted.

Back to watermarking there was a great thread on it in the SH forum, but unfortunately the poster hosted the images on Postimage which are just a dreadful image host and the images have since been removed by them.

Here is what he said:

I recently purchased a 96/24 download of Keith Jarrett's Sleeper from Ponomusic. This is on the ECM label, and ECM is distributed by Universal.

I also have the CD, so I thought I would see I could find some objective evidence of watermarking.

Here's what I did:

- Convert the 96/24 download to 44.1/16 using foobar
- Imported the downscaled download and the cd rip of the same track into Audacity
- Since the tracks had slightly different starting points, manually aligned the two tracks
- Inverted one of the tracks
- Mixed and Rendered the two tracks

Attached is the spectrum analysis for the results. It is not as well defined as the example at http://www.mattmontag.com/images/watermarked_full.png , but that may be because of the down-conversion and manual alignment.

So what do you think? Watermark?

Looks very suspicious to me.

Unfortunately only the tiny thumbnails come up in a Google Image Search:

Here is the thread in question: http://forums.stevehoffman.tv/threa...g-audible-watermarks-on-digital-files.334991/

So many people doubting his tests, refuse to believe watermarking exists and so on.

Well I guess I'm never buying anything there again...

Absolutely. HDTracks will use the defense that they are "just the vendor", but it is their responsibility to inform the labels that they won't stand for it.

 

[maverickronin]

Absolutely. HDTracks will use the defense that they are "just the vendor", but it is their responsibility to inform the labels that they won't stand for it.

...or inform the buyers that it's no longer lossless and adjust their prices accordingly...

 

[andreasmaaan]

...or inform the buyers that it's no longer lossless and adjust their prices accordingly...

Perhaps their argument is that technically nothing’s been lost, something’s just been added

Seriously, it would be good to have this verified by measurement though...

How reliable do you think those little thumbnails that you found are, @hvbias?

 

[hvbias]

Perhaps their argument is that technically nothing’s been lost, something’s just been added

Seriously, it would be good to have this verified by measurement though...

How reliable do you think those little thumbnails that you found are, @hvbias?

When I was able to view the full images I concluded his test was solid. With the inverted/subtracted hi-res vs the CD rip on the spectrograph you could very clearly see "chunking" of audio data in very precise fixed intervals when zoomed way in.

The main issue with this test is you need the hi-res download and CD to have identical mastering (ie mastering engineer mastered to hi-res then created a down sample to 16/44 for the CD, the label then later adds the watermark to the download). You can't just use the hi-res download and any other CD. In my experience this is not very common since the hi-res is usually done by in house mastering people for the sole purpose of the hi-res download. And the CD would have existed before this or gets mastered again at a later date down the road. ECM (and Rush mentioned below) seem to be a special case since they were doing a CD remastering campaign around a similar time HDTracks/Pono were putting out the hi-res downloads.

I have followed the big hi-res discussion thread over there for a while and as far as I know that might have been the only way to view them on spectrals, there was no way to view it with just a known watermarked hi-res download. But I stopped reading that thread as it became a real chore; all subjective opinions, high signal to noise, etc.

I think someone made similar graphs in the Rush reissue thread, but I really don't want to be a masochist and search through 5000+ posts Using Google Image Search I only found the two images I posted above so I might be wrong.
http://forums.stevehoffman.tv/threads/rush-new-2015-vinyl-and-hi-res-reissues-thread-part-2.432219/

 

[hvbias]

Do any of you know what app would have been used to create this spectrograph? I might be onto something more with being able to look at watermarks.

 

[watchnerd]

I'm no expert on this topic, which is why the title of this thread is a question, but it seems clear to me that a significant proportion of music distributed via streaming services is now watermarked by the major labels. Some examples of watermarked audio that have been posted in various places online seem to show that this practice does (or at least can) create quite audible distortion. The examples I've heard seemed to create more objectionable distortion than e.g. that caused by higher bitrate lossy compression, for example.

So I thought it might be worthwhile to try to develop a better understanding of what's going on here.

My questions are:

1. Which labels are doing this and how much of their catalogues is subject to it?
2. What different techniques are being used, and what are the technical and audible differences between these?
3. Is watermarking also being employed in respect of non-streaming releases?

Any solid information would be greatly appreciated.

What do we do now?

Torrents.

 

[mansr]

Do any of you know what app would have been used to create this spectrograph? I might be onto something more with being able to look at watermarks.

Looks like Spek. The clue is in the top right corner.

 

[andreasmaaan]

@svart-hvitt abd i were discussing on another thread the idea of testing a few Tidal lossless/hi-res streams for watermarking and trying to work out, if present, how severely they degraded the signal.

Problem is, I don’t subscribe to Tidal! If anyone does subscribe, and can provide some snippets for comparison, I’d be happy to see if I can get any meaningful data out (no promises I have the skill to do so however).

PM if interested

 

[svart-hvitt]

@svart-hvitt abd i were discussing on another thread the idea of testing a few Tidal lossless/hi-res streams for watermarking and trying to work out, if present, how severely they degraded the signal.

Problem is, I don’t subscribe to Tidal! If anyone does subscribe, and can provide some snippets for comparison, I’d be happy to see if I can get any meaningful data out (no promises I have the skill to do so however).

PM if interested

I have both Tidal and Qobuz. Don’t know how to snip. Tidal has, just checked, 3 months trial for €1. I will owe you a beer for that one