How the Music Industry is Building the Tech to Hunt Down AI-Generated Songs (theverge.com) 38
The goal isn't to stop generative music, but to make it traceable, reports the Verge — "to identify it early, tag it with metadata, and govern how it moves through the system...."
"Detection systems are being embedded across the entire music pipeline: in the tools used to train models, the platforms where songs are uploaded, the databases that license rights, and the algorithms that shape discovery." Platforms like YouTube and [French music streaming service] Deezer have developed internal systems to flag synthetic audio as it's uploaded and shape how it surfaces in search and recommendations. Other music companies — including Audible Magic, Pex, Rightsify, and SoundCloud — are expanding detection, moderation, and attribution features across everything from training datasets to distribution... Vermillio and Musical AI are developing systems to scan finished tracks for synthetic elements and automatically tag them in the metadata. Vermillio's TraceID framework goes deeper by breaking songs into stems — like vocal tone, melodic phrasing, and lyrical patterns — and flagging the specific AI-generated segments, allowing rights holders to detect mimicry at the stem level, even if a new track only borrows parts of an original. The company says its focus isn't takedowns, but proactive licensing and authenticated release... A rights holder or platform can run a finished track through [Vermillo's] TraceID to see if it contains protected elements — and if it does, have the system flag it for licensing before release.
Some companies are going even further upstream to the training data itself. By analyzing what goes into a model, their aim is to estimate how much a generated track borrows from specific artists or songs. That kind of attribution could enable more precise licensing, with royalties based on creative influence instead of post-release disputes...
Deezer has developed internal tools to flag fully AI-generated tracks at upload and reduce their visibility in both algorithmic and editorial recommendations, especially when the content appears spammy. Chief Innovation Officer Aurélien Hérault says that, as of April, those tools were detecting roughly 20 percent of new uploads each day as fully AI-generated — more than double what they saw in January. Tracks identified by the system remain accessible on the platform but are not promoted... Spawning AI's DNTP (Do Not Train Protocol) is pushing detection even earlier — at the dataset level. The opt-out protocol lets artists and rights holders label their work as off-limits for model training.
Thanks to long-time Slashdot reader SonicSpike for sharing the article.
"Detection systems are being embedded across the entire music pipeline: in the tools used to train models, the platforms where songs are uploaded, the databases that license rights, and the algorithms that shape discovery." Platforms like YouTube and [French music streaming service] Deezer have developed internal systems to flag synthetic audio as it's uploaded and shape how it surfaces in search and recommendations. Other music companies — including Audible Magic, Pex, Rightsify, and SoundCloud — are expanding detection, moderation, and attribution features across everything from training datasets to distribution... Vermillio and Musical AI are developing systems to scan finished tracks for synthetic elements and automatically tag them in the metadata. Vermillio's TraceID framework goes deeper by breaking songs into stems — like vocal tone, melodic phrasing, and lyrical patterns — and flagging the specific AI-generated segments, allowing rights holders to detect mimicry at the stem level, even if a new track only borrows parts of an original. The company says its focus isn't takedowns, but proactive licensing and authenticated release... A rights holder or platform can run a finished track through [Vermillo's] TraceID to see if it contains protected elements — and if it does, have the system flag it for licensing before release.
Some companies are going even further upstream to the training data itself. By analyzing what goes into a model, their aim is to estimate how much a generated track borrows from specific artists or songs. That kind of attribution could enable more precise licensing, with royalties based on creative influence instead of post-release disputes...
Deezer has developed internal tools to flag fully AI-generated tracks at upload and reduce their visibility in both algorithmic and editorial recommendations, especially when the content appears spammy. Chief Innovation Officer Aurélien Hérault says that, as of April, those tools were detecting roughly 20 percent of new uploads each day as fully AI-generated — more than double what they saw in January. Tracks identified by the system remain accessible on the platform but are not promoted... Spawning AI's DNTP (Do Not Train Protocol) is pushing detection even earlier — at the dataset level. The opt-out protocol lets artists and rights holders label their work as off-limits for model training.
Thanks to long-time Slashdot reader SonicSpike for sharing the article.
Witch Hunting (Score:4, Interesting)
It won't take long for the first false positive to occur, just look at the mistakes made by YouTube's Content ID system.
This could also lead to a dangerous system where companies (let's be honest, it's not about the artists but about the music industry) trademark certain musical styles and sue anyone who uses them, regardless of whether they knew someone else had used that style before. That's bad news for independent artists, regardless of whether they use AI or not.
Re: (Score:1)
Re: (Score:2)
They may or may not succeed in making this a law, but the article suggests they certainly want to deplatform tracks that have stems similar to those in their portfolio. If YouTube and Spotify cooperate, independent artists will have to pay their tax for having music similar to anything in the vast portfolio of large recording companies. Music isn't that unique. If one has a large portfolio and looks at short segments, one always finds something similar and then can demand to be compensated for that similari
Re: (Score:1)
Re: (Score:2)
It is also not reasonable. But suing people for music playing in the background of their home videos is not reasonable either, nor is wanting kindergartens to license the music they sing.
Re: (Score:2)
That's bad news for independent artists
That's exactly what an AI would say. *suspicious eyes*
Re: (Score:2)
They will find a way to use this on humans. If you do anything fitting their pattern, then you are stealing their influence on you and must pay for the impression left on your brain by them. If you can do the same with a machine why can't you do it with the real thing?
Re: (Score:3)
Re: (Score:1)
Re: Modern Music: No bands. Pitch-correct Compress (Score:1)
Re: (Score:1)
Real music is made by musicians, playing live (Score:3)
AI generated stuff may be fun to make and listen to, and it may be a fad for a short time, but it should be honestly labelled.
The article mostly references AI generated frauds, designed to extract money from streaming services.
For those with no musical skill, tech can allow them to make stuff that almost sounds like music if you aren't too critical. It can be fun to do, but don't expect to get famous or make a lot of money.
I tried for years to use sequencers and virtual instruments to produce my compositions. At best, they sounded OK, but were missing the magic that happens when real musicians play live together.
It's also interesting to observe that when the legendary musical genius Frank Zappa tried using sequencers and virtual instruments, his results also sounded robotic and unmusical. Even when they made small mistakes, his bands always sounded better
Re:Real music is made by musicians, playing live (Score:5, Informative)
What about AI music played live? Google recently released a DJ tool capable of generating music in real-time. The DJ can control how the music changes over the next two seconds, with a 10-second context window. This isn't a tool for creating songs but rather for performing live music.
https://magenta.withgoogle.com... [withgoogle.com]
Re: (Score:1)
Re: (Score:2)
Of course you can record the whole show and release that Live! DVD/CD/ETC. Make it available to rent, digital purchase, dvd or bluray.
Re: (Score:1)
Re: (Score:1)
Imagine the economic chaos if I could just tell my computer what kind of music I want to listen to and it could generate that music for me.
Think of the starving Music Executives!
Tech war (Score:2)
What if a competing AI managed to convince this AI to shut itself down? Or better yet, what if it convinced this AI that these aren't the songs you're looking for and it can go about its business. Move along. Move along.
Re: (Score:2)
That doesn't work as entertaining as you think, but the concept is adversarial noise. You add a crafted noise to the song (if possible unhearable by humans) that detracts the classification from the song being AI. Of course that's a cat and mouse game and rarely as robust as you'd wish, if you look for example at glaze.
Re: (Score:2)
Re: (Score:2)
You're a Youtube manager. The recording industry says "use our new content id system and force AI musicians to buy licenses for our similar music, or we do not license our music to you anymore". What do you answer?
Re: (Score:2)
I'm hoping my bosses will sue them for monopoly abuse. Alphabet should be able to spot one easy.
Why (Score:5, Insightful)
Re: (Score:2)
Re: (Score:3, Informative)
Because The Mushc Industry must be able to continue rent-seeking and collecting most of the revenue that actual musicians create.
I know a guy who was a moderately successful musician in the 90s with a few songs near the top of the charts back then. He loves AI because he no longer needs a band and can create entire songs by himself.
That terrifies The Music Industry. Because its income is based on gatekeeping and rent-seeking.
Re: (Score:2)
They likely want to limit it early on. For two decades, people have been able to self-publish music on the internet and no longer need the music industry. For some reason, this never posed a significant threat to their income or control, as the ways consumers discover music remained largely the same. The challenge was more about how self-publishing musicians could find an audience.
Now a new frontier is emerging, a whole new category of music. This could include not only AI-generated music created by humans
Re: (Score:1)
I think it will be difficult for the music industry to push 'this song was created by an AI in which 0.1% of the training data was a song we own' because that would pretty much destroy the AI industry which is based on being able to use other people's content without paying for it and the Tech billionaires are richer than the music billionaires. You're right though that the tech companies like Youtube might agree to block AI music 'because of unresolved legal questions' or some such nonsense.
Re: (Score:2)
Maybe I'm wrong to equate this, but what is the difference between me creating a song on my own versus me using ai to create a song? Neither one of those interactions actually requires RIAA permission. I can create and record my own music and give it away all I want. So why can't I also use AI to make give away music? See what I'm saying?
I could also make and record my own music and charge people for it. That doesn't require me signing with the RIAA. Sure, discovery may be really hard, but then again, maybe
Re: (Score:2)
Actually, now that I think about it, I just described Myspace, right?
Re: (Score:2)
Does the "music industry" need to hunt down AI generated songs? It's not their songs and it's none of their business where they came from.
Because of music fraud. There's a very real problem right now of fake music being uploaded under real artists names for the purpose of click farming. It's the same as AI generated books that are nothing but incomprehensible trash that pretends to be from real authors to capitalise on their names.
The problem is the bar for checking currently is "is this music yes / no", and beyond that uploads to streaming sites are generally allowed. It's then relies on a manual process to call it out and get it taken down
They don't want the competition (Score:4, Interesting)
Re: (Score:2)
Hunt them down and copy it to replace the artists (Score:1)
I think the RIAA is scared (Score:2)
Creative Influence? (Score:2)
with royalties based on creative influence
If you are a musician you better be careful in interviews about who your influences were. The corporations who own their music are going to come after you.