Google is embedding inaudible watermarks right into its AI generated music::Audio created using Google DeepMind’s AI Lyria model will be watermarked with SynthID to let people identify its AI-generated origins after the fact.
People are listening to AI generated music? Someone on Bluesky put (paraphrased slightly) it best-
If they couldn’t put time into creating it I’m not going to put time into listening to it.
I think I’d rather listen to some custom AI generated music than the same royalty free music over and over again.
In both cases they’re just meant to be used in videos and stuff like that, you’re not supposed to actually listen to them.
Fun fact: Steve1989MREInfo uses all of his original music for his videos.
This is the ultimate YouTuber power move. Exurb1a and RetroGamingNow do it too!
A number of Youtubers do . . . and some of it’s even good, lol. John at Plainly Difficult and Ahti at AT Restorations are two that use their own music that I can think of off the top of my head.
Sam with Geowizard. actually quite a few “big” channels do which is awesome
People are using AI tools to do crazy stuff with music right now. It’s pretty great
Human performance but AI voice: https://www.youtube.com/watch?v=gbbUWU-0GGE
Carl Wheezer covers: https://www.youtube.com/watch?v=65BrEZxZIVQ
Here is an alternative Piped link(s):
https://www.piped.video/watch?v=gbbUWU-0GGE
https://www.piped.video/watch?v=65BrEZxZIVQ
Piped is a privacy-respecting open-source alternative frontend to YouTube.
I’m open-source; check me out at GitHub.
You tell 'em, bot. 🙌🏽
Can it be much different from the mass-market auto-tuned pap that gets put out today?
The singers of that music actually have to use their voice to sing into a mic compared to someone on a computer typing in a prompt.
As much as I dislike modern pop music, I will definitely say they put in more work than the people who rely solely on an AI that will do all the work based on a prompt.
My own feelings on the matter aside (fuck google and all that) this has been something chased after for a long time. The famous composer Raymond Scott dedicated the back end of his life trying to create a machine that did exactly this. Many famous musical creators such as Michael Jackson were fascinated by the machine and wanted to use it. The problem was is he was never “finished”. The machine worked and it could generate music, it’s immensely fascinating in my opinion.
If you want more information in podcast format check out episode 542 of 99% invisible or here https://www.thelastarchive.com/season-4/episode-one-piano-player
They go into the people who opposed Scott and why they did, and also talk about the emotion behind music and the artists, and if it would even work. Because the most fascinating part of it all was that the machine was kind of forgotten and it no longer works. Some currently famous musicians are trying to work together to restore it.
The question then is, if someone created their life’s work and modern musicians spend an immense amount of time restoring the machine, when the machine creates music does that mean no one spent time on it? I enjoy debating the philosophy behind the idea in my head, especially since I have a much more negative view when a modern version of this is done by Google.
I feel like the machine itself would be the art in that case, not necessarily what it creates. Like if someone spent a decade making a machine that could cook FLAWLESS BEEF WELLINGTON, the machine would be far more impressive and artistic than the products it made
i mean, where do you draw the line necessarily between the machine and what it creates? the machine itself is totally useless without inputs and outputs, not to say art needs utility. the beef wellington machine is only notable on its ability to conjure beef wellington, otherwise it’s just a nothing machine. which is still kind of cool, I guess, but the beef wellington machine not making beef wellington is kind of a disregard for the core part of the machine, no?
That was a great episode of 99PI. Would love the machine restored.
IIRC, It’s not so much that it made music, but that it would create loops through iteration to inspire people. He wanted it to make full busic but it was never close to that
Yeah I think you’re right, and it was apparently actually random. The longer it would play a loop the more it would iterate. Such a cool thing to exist
You will still listen to it, watching movies, advertisements, playing video games…
This is the worst timeline
This is the worst time line so far.
Not yet.
Yikes. TIL you think music sounds good based on how much time went into making it, not how it actually sounds.
Can’t wait for you to hear something you like then pretend it’s bad when you find out it was made by AI.
This assumes music is made and enjoyed in a void. It’s entirely reasonable to like music much more if it’s personal to the artist. If an AI writes a song about a very intense and human experience it will never carry the weight of the same song written by a human.
This isn’t like food, where snobs suddenly dislike something as soon as they find out it’s not expensive. Listening to music often has the listener feel a deep connection with the artist, and that connection is entirely void if an algorithm created the entire work in 2 seconds.
What if an AI writes a song about its own experience? Like how people won’t take its music seriously?
It will depend on whether or not we can empathize with its existence. For now, I think almost all people consider AI to be just language learning models and pattern recognition. Not much emotion in that.
just language learning models
That’s because they are just that. Attributing feelings or thought to the LLMs is about as absurd as attributing the same to Microsoft Word. LLMs are computer programs that self optimise to imitate the data they’ve been trained on. I know that ChatGPT is very impressive to the general public and it seems like talking to a computer, but it’s not. The model doesn’t understand what you’re saying, and it doesn’t understand what it is answering. It’s just very good at generating fitting output for given input, because that’s what it has been optimised for.
Glad you’re at least open to the idea.
“I dunno why it’s hard, this anguish–I coddle / Myself too much. My ‘Self’? A large-language-model.”
I noticed some of your comments are disappearing from this thread, is that you or mods?
I’m getting nuked from another thread
Language models dont experience things, so it literally cannot. In the same way an equation doesnt experience the things its variables are intended to represent in the abstract of human understanding.
Calling language models AI is like calling skyscrapers trees. I can sorta get why you could think it makes sense, but it betrays a deep misunderstanding of construction and botany.
What makes your experiences more valid than that of AI?
It is not a measure of validity. It is a lack of capacity.
What is the experience of a chair? Of a cup? A drill? Do you believe motors experience, while they spin?
Language models arent actual thought. This isnt a discussion about if non organic thought is equivalent to organic thought. Its an equation, that uses words and the written rules of syntax instead of numbers. Its not thinking, its a calculator.
The only reason you think a language model can experience is because a marketing man missttributed it the name “AI.” Its not artificial intelligence. Its a word equation.
You know how we get all these fun and funny memes where you rephrase a question, and you get a “rule breaking” answer? Thats because its an equation, and different inputs avoid parts of the calculation. Thought doesnt work that way.
I get that the calculator is very good at calculating words. But thats all it is. A calculator.
What makes your thoughts ‘actual thought’ and not just an algorithm?
Oddly, I’d find a piece of music written by an ai convinced it was a chair extremely artistic lol. But yeah, just because the algorithm that’s really good at putting words together is trying to convince you it has feelings, doesn’t mean it does.
That’s a parasocial relationship and it’s not healthy, sure Taylor Swift is kinda expressing her emotions from real failed relationships but you’re not living her life and you never will. Clinging to the fantasy of being her feels good and makes her music feel special to you but it’s just fantasy.
Personally I think it would be far better if half the music was ai and people had to actually think if what their listing to actually sounds good and interesting rather than being meaningless mush pumped out by an image obsessed Scandinavian metal nerd or a pastiche of borrowed riffs thrown together by a drug frazzled brummie.
Lol, somehow you got the above commenter covering the sentiment that a song is better if it’s message is true to its creator…something a huge percentage of the population would agree with, and you equate that to fan obsession.
People on the internet are wild.
I don’t understand where they got any of that from, lol. It’s like they learned what a parasocial relationship is earlier today and they thought it applied here
I would kind of agree with this if it wasn’t kind of mean and half of it didn’t come out of nowhere, but then it also seems like what you think you value in your own music taste is whether or not something is new, seeing as your main examples of things that are meaningless or bad is “image obsessed scandinavian metal nerd” i.e. derivative and “pastiche of borrowed riffs thrown together by a drug frazzled brummie” i.e. derivative.
Ha no you are right, I was being a dick - i worked long enough in the music industry that it’s scarred my soul and just thinking of it brings up that bile…
But yeah I was just being silly with the band descriptions, I was describing some of the music I like in a flippant way to highlight the absurdity of claiming some great artistic value because Ozzy mumbled about iron man traveling time for the future of mankind - dice could come up with more meaningful lyrics than ‘nobody helps him, now he has his revenge’ is the sort of thing an edgy teenage coke head would come up with – it’s one of my favourite songs of all time, another example of greatest songs of all time is Rasputin by boney m, famously part of a big controversy when people discovered they were a manufactured band and again the lyrics and music are both brilliant and awful.
People obsess over nonsence all the time, it’s easy to pretend there’s some deep and holy difference between Bach and Offenbach but the cancan can mean just as much as any toccata if you let it.
Art is in the eye of the beholder, it has always been thus and will always be thus.
deleted by creator
I can’t tell if you’re completely missing the point on purpose, or if you actually don’t understand what I mean lol. Who said anything about Taylor Swift?
You can replace her with whatever music you associate with, what I’m getting at is your connection to it isn’t real - it feels real but that’s because it’s coming from you, you’re putting the meaning in there.
If you could erase all memory of Bach from a classical obsessives mind then play them his greatest hits and say it’s from an AI they’d say ‘ugly key smashing meaningless drivel’ maybe they’d admit AI Brahms has some bangers but without the story behind it and the history of its significance it’s not as magical.
The problem I have is people are to addicted to shortcuts, ‘oh this is Bach people say he’s great so this cello suite must be good therefore I like it’ it’s lazy and dumb. (I use Bachs cello suite as an example because it’s what’s on the radio but you can put any bit of music as the example)
I don’t think that’s OPs point, but it’s interesting how many classic songs were written in less than 30 minutes
deleted by creator
deleted by creator
deleted by creator
That’s not really a gotcha though. They’re saying they aren’t going to actively seek out and listen to auto-generated music. If they happen to hear some and like it, that wouldn’t mean they actively sought it out and listened to it.
I’m not going to put time into listening to it.
Right, they’re not going to actively put time into listening to music generated by AI.
Hearing music made by AI because it happens to be playing is different from knowingly listening to it. It’s alarming that you need this spelled out so much.
Don’t know why you’re being downvoted, you’re completely right. I’d never seek out to listen to something with no human thought process behind it
Ok, boomer.
How’s that microwave dinner taste? Like an A for effort? Yeah, I bet.
deleted by creator
deleted by creator
A spectrum analysis and bandpass filter should take care of that.
chuckles contemptfully in Audacity
So we’ll just need another AI to remove the watermarks… which I think already exists.
Don’t even need AI. Basic audio editing works.
The Audacity!
Hehe.
This is the best summary I could come up with:
Audio created using Google DeepMind’s AI Lyria model, such as tracks made with YouTube’s new audio generation features, will be watermarked with SynthID to let people identify their AI-generated origins after the fact.
In a blog post, DeepMind said the watermark shouldn’t be detectable by the human ear and “doesn’t compromise the listening experience,” and added that it should still be detectable even if an audio track is compressed, sped up or down, or has extra noise added.
President Joe Biden’s executive order on artificial intelligence, for example, calls for a new set of government-led standards for watermarking AI-generated content.
According to DeepMind, SynthID’s audio implementation works by “converting the audio wave into a two-dimensional visualization that shows how the spectrum of frequencies in a sound evolves over time.” It claims the approach is “unlike anything that exists today.”
The news that Google is embedding the watermarking feature into AI-generated audio comes just a few short months after the company released SynthID in beta for images created by Imagen on Google Cloud’s Vertex AI.
The watermark is resistant to editing like cropping or resizing, although DeepMind cautioned that it’s not foolproof against “extreme image manipulations.”
The original article contains 230 words, the summary contains 195 words. Saved 15%. I’m a bot and I’m open source!
it does this by converting the audio into a 2d visualisation that shows how the spectrum of frequencies evolves in a sound over time
Old school windows media player has entered the chat
Seriously fuck off with this jargon, it doesn’t explain anything
That’s actually an accurate description of what is happening: an audio file turned into a 2d image with the x axis being time, the y axis being frequency and color being amplitude.
That’s literally a spectrograph
Spectrogram*
Your mom’s literally a spectrograph.
I know, it’s like the old windows media player visualisations.
Sounds like a bad journalist hasn’t understood the explanation. A spectrogram contains all the same data as was originally encoded. I guess all it means is that the watermark is applied in the frequency domain.
Also this isn’t new by any stretch… Aphex Twin would like a word
Well, encoding stuff in the spectrogram isn’t new, sure. But encoding stuff into an audio file that is inaudible but robust to incidental modifications to the file is much harder. Aphex Twin’s stuff is audible!
I would like to know what it is that makes it so robust. The article explains very little. Is it in the high frequencies? Higher than the human ear can hear? Compression will effect that plus that’s going to piss dogs off. Could be something with the phasing too. Filters and effects might be able to get rid of the water mark
I don’t know what frequencies are annoying for dogs but I’m guessing it’s above 24kHz so no sound file or sound system is going to be able to store or produce it anyway.
There will certainly be some way to get rid of the watermark. But it might nevertheless persist through common filters.
thats like putting a watermark besides the Bill.if it is inaudible then you can just delete it
So basically it’s security through obscurity, since once people know they can and will edit it out, especially those who want to use it for deception.
