If AI replaces musicians, does the entire plugin industry die with them?

[vurt]

did we ever get jesus riding a lobster?

[jamcat]

But it's probably around 25% of the global music market. If you can't break the US market, you're never going to reach the really big time. That's just a fact.

I don't think that's true. Do you have a source for that "fact"?

Here you go.

https://www.billboard.com/pro/riaa-2025 ... streaming/

I have tons more I can ram down your throat.

I didn't reply to you but to "BONES". It's interesting you answered though, and you both also have your "gear" listed in your signatures.

By jove, you've done it! BONES and Wags are the same person! It all makes sense now why their music is so eerily similar...

[guitarzan]

Today for my Gemini/Lyria test I tried just a basic Chicago blues guitar jam. I always try to keep vocals out of the generated clip because I am evaluating only the instrumental performance at this time.

I am trying to see if Gemini/Lyria can sound like it’s jamming, I am basically holding auditions for my personal virtual jam band. I think it holds up well for AI . It knows how to at least attempt to try to sound pushed and not so session perfect like most AI music generators seem to be always trying to produce.

Gemini/Lyria has done a great job with every instrument I’ve tried at sounding pushed — except banjo, and I was looking for the opposite there and it passed with flying colors. With banjo, even a lot of the top bluegrass players NEVER let up, it’s like they think full out is the only way to play banjo, so when I did banjos I was looking for a controlled touch as that is what it takes to play banjo with any dynamic range.

Anyway, here’s the little blues thing. The Supro amp is a nice touch:

https://g.co/gemini/share/5d158d6911fb

Edit: Gemini/Lyria actually knows what amp sag is! Once considered mythical by amp sim developers — I feel like I’ve made a difference!!

[Zeisner]

It's not actually clear if the compression artifacts are from lossy training data, or from compression Suno and or Youtube are using.

Dude. I worked in that field. In assisted scientists in psychoacoustic research using machine learning. You're only proving that you can't hear anything because those AI artifacts I'm referring to sound very different from lossy compression codecs - which don't add pseudorandom amplitude/pitch/timbre fluctuations because they work with particular window sizes (Old lossy codecs are a good example, their compression artifacts are much more audible). The difference is HUGE. It's like the difference between red dots on a white background vs a yellow-blue checkerboard (to use a visual metaphor). If you try to tell me that those are the same patterns and colors then the only conclusion I can come to is that you're seriously impaired.

But there is no such thing as "composition artifacts."

There are always artifacts in everything because this is how machine learning works. There is no boolean logic or math. To refuce the amount of artifacts to a meaningful level (0,001 % and below depending on the subject) you need an awful amount of high-quality data. Neither Suno nor Tunee nor any other generative music AI company has that because the data simply doesn't exist. There have been complains about generated songs sounding too much like mainstream pop when they're supposed to sound like a non-pop/underground genre. That's an artifact caused by faulty and missing training data and it affects every aspect from composition to the mix.

And no, I wasn't attacking you. I genuinely think that your hearing and understanding of music and audio production must be really bad if you consider those examples catchy.

[Zeisner]

did we ever get jesus riding a lobster?

You can find it on YouTube.

[Auplant]

It might make formulaic music, and well known genres. But I'm doubtful that it could make good experimental music, and good original ambient music (with deep human emotion) when it is still a psychopath. I don't want to pay Suno to find out.

[bermudagold]

And this 100% AI song reached #1 on the Billboard Country charts:

It's not an international chart though? Only in the USA and even there I guess it's only certain parts of the USA. The USA population is like ~4% of the world population.

The USA is the largest market for music transactions in the world. There are only 3 EU countries in the top 10...so if we are talking monetizing of music and maximizing revenue, ...the EU doesn't matter as much as you think...asia matters more...on top of that the largest and most successful PROs (Performers Rights Organizations) are based in the US...these are the orgs that chase down, measure, determine, and collect royalties for publishing owners to get people paid...for example Adele is a UK artist, but she is signed with a USA PRO...Sesac

Why are you talking about the EU though? The EU is not the whole of Europe nor along with the USA, the whole world.

I don't think anybody said either of those things...simply the fact that the USA regardless of its percentage of global population dominates virtually every metric of the music industry...so its data is used as the benchmark

[jamcat]

There have been complains about generated songs sounding too much like mainstream pop when they're supposed to sound like a non-pop/underground genre. That's an artifact caused by faulty and missing training data and it affects every aspect from composition to the mix.

And no, I wasn't attacking you. I genuinely think that your hearing and understanding of music and audio production must be really bad if you consider those examples catchy.

So which is it?
Does AI make music that that sounds like mainstream pop or does it make music that isn't catchy? Because it can't be both.

[bermudagold]

Today for my Gemini/Lyria test I tried just a basic Chicago blues guitar jam.

I am trying to see if Gemini/Lyria can sound like it’s jamming, I am basically holding auditions for my personal virtual jam band. I think it holds up well for AI . I

Anyway, here’s the little blues thing. The Supro amp is a nice touch:

https://g.co/gemini/share/5d158d6911fb

honestly, I doubt the average spotify subscriber or even the almost extinct album buyer, could tell that's AI

[Zeisner]

But I'm doubtful that it could make good experimental music

Of course it can't. It's just a stochastic parrot. It has no feelings, no senses, no thoughts. Simulating all of that is out of reach because scientists still struggle to even understand how neural networks in a human brain works.

To use a metaphor: the smartest people on Earth struggle with designing a car door that stays closed while other people fantasize about building cars which could fly to the Moon and back.

[guitarzan]

It might make formulaic music, and well known genres. But I'm doubtful that it could make good experimental music, and good original ambient music (with deep human emotion) when it is still a psychopath. I don't want to pay Suno to find out.

I don’t know anything about ambient music, and this is Gemini/Lyria. I chose this one because it’s sometimes a struggle to get Gemini to leave out its cheesy lyrics (Beta thing, I’m trying to make constructive suggestions):

https://g.co/gemini/share/c3d7003e09d3

[Zeisner]

Does AI make music that that sounds like mainstream pop or does it make music that isn't catchy? Because it can't be both.

As if all mainstream pop would be catchy. Once again you're proving that you don't understand music.

[jamcat]

"Pop" is short for "popular." Being liked by a wide swath of people is literally the only definition of pop. Being catchy is what makes it popular. Perhaps you are the one who doesn't "understand music."

[BONES]

Jamcat, for the win!

I don't think that's true. Do you have a source for that "fact"?

The fact was the second bit, the figures were a guess. You can check the actual facts HERE, which aren't too far off my guess. Interestingly, though, there are no figures for the last few years.

If it worked it could have been useful for research or to generate samples of instruments I don't have access to. But this is a joke

No, mate, you're a joke. AI is not set-up to do that stuff but that doesn't mean it has nothing to offer anyone. It's just you who misses out by being a pedant with absurdly unrealistic expectations.

I'm baffled how anyone could think that any of this sounds good enough for anything aside from playing musical noise in the background of a café and even that is debatable.

I'm baffled that anyone thinks that's important in this era of mp3s, Bluetooth and ear buds. Honestly, if those things are a factor, then the music you're listening to can't have a whole lot going for it.

But on the other hand - there were plenty of people back then who said that lowpoly models in a N64 game looked totally real...

Nobody I know.

Not a criteria, I was just pointing out country music is really not a thing anywhere else than certain parts of the USA. Or is this AI talk only about the USA?

Are you kidding? Country is massively popular in Australia. Artists like Slim Dusty and Casey Chambers have had incredible mainstream success. And where do you think Keith Urban comes from?

Either you can't hear anything (from composition to mix) or you're trolling. Those are the only two possibilities. Pick your poison.

I definitely can't hear anything in that metal guitar example. The AI guitar sounds just as good as the real guitar to my ears, maybe even better. But it is compressed YouTube audio, so it's hardly definitive, but it is indicative of a typical listening situation.

I can hear artifacts and aliasing in some of my VSTi but it's just part of their character. It doesn't worry me in the slightest and it's mostly drowned out by the mix anyway. You're tilting at windmills.

... you both also have your "gear" listed in your signatures. And you're not gonna "ram" anything "down my throat".

Do you not understand the benefit of listing your gear? Think about it for a minute and see what you can come up with. I'll give you a hint - it's pretty standard practice for beta-testers on a company forum.

those AI artifacts I'm referring to sound very different from lossy compression codecs... The difference is HUGE.

So point some out to us, please. If it's so completely obvious, it shouldn't be hard for us to pick it up if you point us in the right direction. Unlike you, I am always willing to learn something.

That said, you have to see that telling us it's not like compression artifacts kind of shoots your argument about training on compressed music down in flames.

To refuce the amount of artifacts to a meaningful level (0,001 % and below depending on the subject) you need an awful amount of high-quality data.

That seems like a rather arbitrary standard, why is that a relevant number?

Neither Suno nor Tunee nor any other generative music AI company has that because the data simply doesn't exist.

Is this you admitting it's all in your head?

There have been complains about generated songs sounding too much like mainstream pop when they're supposed to sound like a non-pop/underground genre. That's an artifact caused by faulty and missing training data and it affects every aspect from composition to the mix.

That's not an artifact, it's the AI not paying attention to the prompt. If you look at AI's like Suno, you can tell it how much weight to place on the prompt. The default is just 25%, so obviously it's not always going to do what you've asked unless you bump it up. I'd suggest people complaining probably don't know what they are doing.

It might make formulaic music, and well known genres. But I'm doubtful that it could make good experimental music, and good original ambient music (with deep human emotion) when it is still a psychopath. I don't want to pay Suno to find out.

Good news, you can use Suno for free or you can just search YouTube for examples that will prove you dead wrong. AI can do great Industrial music and even obscure sub-genres like Noise. Ambient is an absolute doddle for it. And you don't have to worry about expression or emotional connection, it's really good at that stuff.

[jamcat]

I have a cowboy hat I got in Australia. It's made out of kangaroo. Also, Australia has a very large beef industry. Tasmanian wagu is quite good. Australia is basically the Texas of the southern hemisphere. So of course country is going to be huge there.