Re: Actions speak louder than words... or do they?
Tbh, I don't think deliberately using an audio watermark is a good idea.
I read something a while back which detailed an algorithm for making inaudible changes to a sound as a form of watermarking. Not inserting tones, but from what I could tell, it used weird filters on it which were supposed to be hard to hear, but easy to see with spectral analysis or some other non-audible method. And even that had issues. I listened to some examples and its effectiveness was really hit or miss. Sometimes I couldn't tell which was unprocessed and which was, other times it was blatantly obvious, and the artifacts were downright annoying. I'll have to find the site if anyone is interested.
In any case, I'm not at all a fan of the inaudible tone idea for multiple reasons. Put simply, it's naive. My reasons for opposing it:
1. Those tones can limit the dynamic range/maximum loudness of a sound, and that'll only get worse if you start mixing multiple sounds together as you'll be subjecting yourself, and your sound card, to layers upon layers of inaudible crud. If it gets bad enough, the limiting built into Windows or your sound card (not sure which, but most modern PCs have something) will kick in and reduce the volume of things. If not that, then you're potentially subjecting your speakers and headphones to harm; they weren't meant to take that kind of beating after all. In moderation it's okay; everything has some infrasonic and ultrasonic content, and careful use of isolated tones aren't going to hurt anything, but blaring loud stacked explosions and gunfire sounds which have all been layered with equally loud but inaudible tones is not something speaker and headphone designers anticipate, so far as I know. If nothing else, that'll just create audible distortion and physical discomfort (discomfort resulting from exposing someone to loud tones at the very edge of their hearing is a thing and I'm positive it would lead to hearing loss).
2. Most audio processing algorithms aren't meant to handle those kinds of tones. Some will take it like a champ, if they aren't flawed that is, but some, especially the default resampling/pitch shifting which comes with Direct Sound, isn't up for that. Try pitching a very high frequency tone around in BGT and it, well, is absolutely horrendous. You can clearly hear aliasing all over the place, at least on my system, and most others (my system has a standard Realtek sound card).
3. If you want to loop your sounds, be prepared to deal with clicks in your loops once you start inserting those watermarks. I mean, there are workarounds (fading the tone in and out at the ends of the file for instance), but it's just extra work... and if the reputation around here is laziness, I don't have the confidence that such work would get done, unless someone was truly dedicated. And someone who was truly dedicated to watermarking would, tbf, probably try some sort of data watermarking first.
4. Those kinds of tones are going to get filtered out by mp3, ogg, and most other lossy algorithms, especially on the high frequency side. Most codecs will very gladly throw away everything above 20 khz at anything below 192 kbps. And once you start getting below 20 khz, you risk getting into the majority's audible range. Given the young average age of people on here, I suspect a large number of them are physically capable of hearing a 20 khz tone... enough to put a dent in the audience. Now whether they'd catch it in practice is another story. Maybe not, maybe the tones would be masked out well, or maybe they play on crappy speakers or headphones which makes those tones impossible to pick up, but of course there's no way to tell how many people would or would not hear it. So if you want ot play it safe and extend to 22 or 24 khz or something that the average person can't perceive, kiss those compact 128 kbps assets good-bye. Even if the encoding preserves them, that's only half the battle; Mp3 for instance has a ton of trouble with stuff above 16 khz. Opus will code things up to 20 khz at most rates, however it has mechanisms to represent high frequency data more coarsely, so you could very well end up hearing the previously inaudible tone being spread into the audible range. So yeah, you might as well just not use lossy compression at that point; it's not designed for this task and so isn't reliable.
5. Finally, assuming that this actually catches on, I am not very optimistic that it would provide benefit. Filters can easily remove inaudible frequencies; they're used in mastering. So, removing the watermarking wouldn't be that difficult with a little detective work, and by a little detective work, I mean probably less than 5 minutes if you're savvy with filters. All you have to do is find out what frequencies are used for watermarking and notch them out. Of course you could make that harder by using a more complex watermark fingerprint such as a modulating tone, narrowband noise etc. But that's not helping much either; a brickwall low cut at 50 hz and brickwall high cut at 16 khz or thereabouts will filter all the inaudible crud nicely regardless of what it is. Furthermore, since we don't have the advantages of spectral analysis, the presence of the watermarks would be very difficult to spot in some situations, particularly on very short sounds, or very loud sounds. And then, what if two people used the same or similar watermark? What if John uses 20,000 hz and Joe uses 20,073 because it's a nice random number? Who will be crazy enough to get down and dirty and figure out what's really going on there? A nerd like me, maybe. But I don't make a habit of that kind of stuff.
So yeah, there are much smarter ways to secure your assets. I don't know what, but there have to be better ways.
Boy I miss the days of the trust/honor system. I know I know, we're long past the days when that actually works, tbh it probably never works really but still. No matter what, though, let's not resort to this form of watermarking. It's one thing to watermark for your own identity and protection to spot the bad apples in the barrel. I get that. But it's completely another thing when you guarantee compromises in the end product because you're that worried about detecting when someone has used something they shouldn't. And if you're naively trying this without being aware of possible side effects, that's even worse.
-- Audiogames-reflector mailing list Audiogames-reflector@sabahattin-gucukoglu.com https://sabahattin-gucukoglu.com/cgi-bin/mailman/listinfo/audiogames-reflector
