'AI Music Generators Could Be a Boon For Artists - But Also Problematic' (techcrunch.com) 48
"Our new robot overlords are making a whole lot of progress in the space of AI music generation," quips TechCrunch, discussing a new project called "Harmonai" backed by Stability AI (creators of the open source AI image generator Stable Diffusion):
In late September, Harmonai released Dance Diffusion, an algorithm and set of tools that can generate clips of music by training on hundreds of hours of existing songs.... Dance Diffusion remains in the testing stages — at present, the system can only generate clips a few seconds long. But the early results provide a tantalizing glimpse at what could be the future of music creation, while at the same time raising questions about the potential impact on artists....
Google's AudioLM, detailed for the first time earlier this week, shows... an uncanny ability to generate piano music given a short snippet of playing. But it hasn't been open sourced. Dance Diffusion aims to overcome the limitations of previous open source tools by borrowing technology from image generators such as Stable Diffusion. The system is what's known as a diffusion model, which generates new data (e.g., songs) by learning how to destroy and recover many existing samples of data. As it's fed the existing samples — say, the entire Smashing Pumpkins discography — the model gets better at recovering all the data it had previously destroyed to create new works....
It's not the most intuitive idea. But as DALL-E 2, Stable Diffusion and other such systems have shown, the results can be remarkably realistic.
Its lyrics are gibberish, TechCrunch concedes — though their article also features several audio clips (including a style transfer of Smash Mouth's vocals onto the Tetris theme).
And the article also notes a new tool letting artists opt of of being used in AI training sets, before raising the obvious concern...
The project's lead stresses that "All of the models that are officially being released as part of Dance Diffusion are trained on public domain data, Creative Commons-licensed data and data contributed by artists in the community." But even with that, TechCrunch notes that "Assuming Dance Diffusion one day reaches the point where it can generate coherent whole songs, it seems inevitable that major ethical and legal issues will come to the fore."
For example, beyond the question of whether "training" is itself a copyright violation, there's the possibility that the algorithm might accidentally duplicate a copyrighted melody...
Google's AudioLM, detailed for the first time earlier this week, shows... an uncanny ability to generate piano music given a short snippet of playing. But it hasn't been open sourced. Dance Diffusion aims to overcome the limitations of previous open source tools by borrowing technology from image generators such as Stable Diffusion. The system is what's known as a diffusion model, which generates new data (e.g., songs) by learning how to destroy and recover many existing samples of data. As it's fed the existing samples — say, the entire Smashing Pumpkins discography — the model gets better at recovering all the data it had previously destroyed to create new works....
It's not the most intuitive idea. But as DALL-E 2, Stable Diffusion and other such systems have shown, the results can be remarkably realistic.
Its lyrics are gibberish, TechCrunch concedes — though their article also features several audio clips (including a style transfer of Smash Mouth's vocals onto the Tetris theme).
And the article also notes a new tool letting artists opt of of being used in AI training sets, before raising the obvious concern...
The project's lead stresses that "All of the models that are officially being released as part of Dance Diffusion are trained on public domain data, Creative Commons-licensed data and data contributed by artists in the community." But even with that, TechCrunch notes that "Assuming Dance Diffusion one day reaches the point where it can generate coherent whole songs, it seems inevitable that major ethical and legal issues will come to the fore."
For example, beyond the question of whether "training" is itself a copyright violation, there's the possibility that the algorithm might accidentally duplicate a copyrighted melody...
Oh goody (Score:4, Insightful)
More repetitive schlock. Nothing like hearing the same melodies (such that they are) from a myriad of "artists".
Music, like writing, should be inspirational. Yes, there are only so many notes (and words), but it's how you arrange them that matters.
Re: (Score:3)
They should be called "creators", not "artist". An artist is someone who produces works of art using talent or inspiration. A creator makes content. There's a slight difference that Youtubers have perfectly captured by choosing - probably by mistake - the correct word to describe themselves (and, ironically, proudly so).
95% of today's music makers create Youtube-worthy content and should therefore be called "creators". Particularly if the content they produce is AI-generated.
Re: (Score:3)
95% of everything has and always will be crap. It's nothing new.
Re: (Score:3)
95% of everything has and always will be crap. It's nothing new.
Also known as Sturgeon's law [wikipedia.org].
Re: (Score:2)
More repetitive schlock. Nothing like hearing the same melodies (such that they are) from a myriad of "artists".
Music, like writing, should be inspirational. Yes, there are only so many notes (and words), but it's how you arrange them that matters.
If arrangement is the problem, then the music making software could be programmed to produce the sequences that most people would find inspirational. Social media companies are notorious for doing something similar to produce user "engagement". The music sounds like "repetitive schlock" because it's early days yet.
Incidentally, "inspirational" music is a fairly recent phenomenon, probably due to the Romantic movement led by composers like Beethoven and Chopin. There were folk songs, yes, but because they we
De do do do, de da da da (Score:1)
Re: (Score:2)
It should work great for Jamiroquai
Re: (Score:2)
How hard can it be to write an AI that uses the chord progression I-V-vi-IV really (what like 90% of pop and country is)? If you don't know what that means, the common piano root of C major is I, G major is V, A minor is vi (lowercase indicates minor), and F major is IV. If it is your Fight Song or if you're Torn about it, you know this progression (or anything by Nickelback, Taylor Swift, P!nk, etc.).
It can't possibly be worse... (Score:3)
...than current pop music.
If AI music is derivative, well, is pop, is rap, is most EDM not derivative (like, to 11)?
Re: (Score:3, Insightful)
Rap isn't derivative. There is only ever a single rap "song", it just gets sampled a bit differently. Rappers dub over a bit of lyrics and call it a day.
Re:It can't possibly be worse... (Score:4, Insightful)
Lol wow... that this gets +5 insightful instead of funny...
Re: (Score:2)
Why don't we have both?
Re: (Score:2)
Good music died with autotune. (Score:3)
Re:Good music died with autotune. (Score:4, Informative)
Automated beats won't help the situation.
This guy has a nice video [youtube.com] (a bit long) about how to tell if someone has used autotune. Fortunately, he uses Freddie Mercury as the non-autotune reference.
One of the comments also has a comment from Freddie who said: "We are in a golden age of music. There will be a time when technology becomes so advanced that we'll rely on it to make music rather than raw talent...and music will lose its soul."
Re: (Score:2)
Freddy Mercury's music is compressed to hell.
Pop music is and has always been shit. Autotune doesn't change that.
Re: Good music died with autotune. (Score:2)
compressed to hell.
Queen's music was among the last ones standing that wasn't compressed to hell. That shit started in the 90s.
Open a Queen song, something from A Night at the Opera for example, in Audacity and look in wonder at the dynamic range. It (almost) justifies getting the 24bit remasters. Almost, but that's another discussion entirely.
Re: (Score:2)
Queen does use dynamic range more than other groups, but they still compress the individual parts.
Re: Good music died with autotune. (Score:2)
Re: (Score:2)
I don't. Autotune makes artists sound better. For example, they actually sing in tune for a change.
They will flood the zone (Score:2)
Re: (Score:2)
And they'll have to take on the guy who copyrighted every possible melody of 10 notes in length or less, and has them all recorded to a hard drive. He has promised to use his copyrights only to counter abuses by others.
Re: They will flood the zone (Score:2)
Re: They will flood the zone (Score:2)
Dunno about 13, but there are many other scales. You're probably familiar with pentatonic (the black keys). Debussy wrote in whole tone scales. Experimental musician Harry Parch (sp?) made a 41-note microtonal scale.
You prime number guys. How's that supposed to work?
Re: They will flood the zone (Score:2)
Re: They will flood the zone (Score:2)
Yeah the guitarist from 75 Dollar Bill has a custom instrument that fuses western scales with stuff that's common in West (I think) Africa. Cool stuff.
About time... (Score:4, Interesting)
The tune had been haunting London for weeks past. It was one of countless similar songs published for the benefit of the proles by a sub-section of the Music Department. The words of these songs were composed without any human intervention whatever on an instrument known as a versificator.
"1984", G. Orwell.
Re: (Score:2)
mod this up. Orwell had a good insight here.
Re: About time... (Score:2)
oh great just what the world needs (Score:2)
Any day now (Score:2)
..there will be the first AI sued for plagiarism.
plagiarism? (Score:1)
Did you know that plagiarism isn't illegal [wikipedia.org].
Good enough & novel music, angry musicians. (Score:1)
Re: (Score:2)
Stable Diffusion can make some amazingly novel art
That's simply not true. Everything Stable Diffusion produces is necessarily similar to the training data. It can produce nothing else. It is incapable of novelty.
on an extremely wide variety of material
Let's bring that back down to earth. The model can contain no more information from the training data than can be stored in the weights. Stable diffusion has just 890 million parameters and a training set of 2.3 billion images. Do the math on that real quick. What "sticks" will be what is most similar to other images. If you play with one of
When will AI start generating idealized porn? (Score:1)
I'm waiting. Kleenex and lotion and all...
Not copyright (Score:2)
For example, beyond the question of whether "training" is itself a copyright violation, there's the possibility that the algorithm might accidentally duplicate a copyrighted melody...
A human could also happen to duplicate an existing melody by chance. It happens all the time. And it's not a problem. Copyright only protects against copying, not independent creation.
Re: (Score:2)
It can be a problem for the second musician when they have to prove that they came up with the tune independently.
Just another tool in the toolbox (Score:1)
Re: (Score:2)
This is yet another step in the evolution of tool use.
This is not just another tool.
You see, this tool is not for artists, it's for publishers.
This tool allows publishers to generate a work based off of their back catalog with a push of a button.
Re: (Score:1)
This is yet another step in the evolution of tool use.
This is not just another tool. You see, this tool is not for artists, it's for publishers. This tool allows publishers to generate a work based off of their back catalog with a push of a button.
I disagree - it absolutely *is* another tool. A tool can be used by anyone who wishes to wield it: artists, publishers, the audience itself. For example, I am not much of a musician - I do poetry myself. I don't have a lot of money to pay someone to illustrate a collection of those poems, or put background music to them. These tools give me what I cannot (or at times, do not wish) to pay for. Yes, a publisher can generate work based off their back catalog - that is a concern, but quite frankly, it is not
Re: (Score:2)
I do poetry myself. I don't have a lot of money to pay someone to illustrate a collection of those poems, or put background music to them. These tools give me what I cannot (or at times, do not wish) to pay for.
...and thereby depriving an illustrator and a composer of work.
Meanwhile this tool enables many more people to do what you do. Even more to the point, AI allows for creation of poems which means that this development will put a lot of your peers, possibly even you, out of a creative job. At the very least it will create a huge amount of competition that will severely reduce your opportunity to make a living off of your work.
I mean, do you really think the market will limit themselves to illustrations and ba
Re: (Score:1)
...and thereby depriving an illustrator and a composer of work.
Up until now that work would not exist if the person who wanted it could not (or would not) pay the market's price for it.
As of now that work has the potential to exist, in ever-increasing (and varying) quantity and quality.
At the very least it will create a huge amount of competition that will severely reduce your opportunity to make a living off of your work. I mean, do you really think the market will limit themselves to illustrations and background music and spare the poor old poet? Think again.
Whether it will spare anyone is irrelevant. What matters is how human creators choose to respond to the genie when it's out of the bottle. IMO, those that can use the technology to create more innovative works are the ones that will fare better. That includes poor old poets :)
Meanwhile, the publishers are the ones who have a serious back catalog that allows them to create AIs that poop out good enough shit that is acceptable to the public at large. These are the only entities that will truly benefit from this development.
That i
Re: (Score:2)
What I disagree with is the assumption that no one else, including those human artists, will benefit from that same technology.
Well, some will obviously profit from AI. But those some will be able to fill a much larger place in the market.
And, in fact, artists as such are not really needed that much in such a world. A computer geek in the IT department could generate the same stuff as artists make. And ultimately the user will just state what he/she wants and a computer will generate it. No human intermediaries at all.
Ai does not hold copyright (Score:2)
As a performer you would still have some protection but the works would not be copyrighted.
Boring (Score:2)
Re: (Score:2)