Just wait until the rush of commenters that insist you’re wrong about being offended that they want to automate and commodify every aspect of your life.
Made this point elsewhere, but the music industry has been on a downward slide towards making music as cheaply as possible for some time now.
If you look at Taylor Swift's first 12 number one hits, each of them was written by a different writer. Compare that to bands from 30 years ago, many of whom wrote and recorded all the songs themselves.
Labels don't sign rock bands anymore because actually recording a rock band well in a studio is 10x the cost of just using a sampler and a single artist singing. I know folks want to blame AI, but it's really just enabling the latest iteration of this trend.
I'm not defending the whole thing. It's a shame, and I love going back and listening to my old Rush albums. But AI is not the problem here. It's the incentives.
I don't have any special industry experience. I went on a tear a few weeks ago and watched a bunch of Rick Beato's videos on YouTube. I didn't know the guy before running across his work, but he has 5 million subscribers and he sure sounds like he knows what he's talking about. He's been a music producer for 30 years.
Anyway, he was the one that made the point that we don't sign rock bands anymore in the sense that they're not moving the industry. All you gotta do is look at the top songs that folks are listening to on Spotify or the radio and you'll immediately see what I'm talking about.
He was also the one that walked through the process of setting up mics for a drum kit and pointed out that it's just very expensive to get the studio time and the expertise to do all that correctly. He actually walks you through a studio where he's set up mics for a drum kit and explains why it's so difficult to do well. He then contrasts that with simply using samples that are professionally provided and that the cost difference is just immense.
Anyway, I don't need to die on this hill. My point was the music industry is going downhill regardless and AI is just one of many tools paving the way.
I agree 100% that mic-ing live drums is by far the hardest and most expensive element in rock recording
But in the 1970s-2000s it was complete black magic and without dedicating years to the craft - you were up to the whims of studios for how much you pay
Compare that today, for instance have a look at the Jazz-Rock Fusion band Vulfpeck’s first album. If you exclude the cost of instruments - they often only need three (rather cheap) mics. Everything else DI. Recorded in a basement for less than a couple grand - with effectively infinite recording time
Live drums are expensive compared to samples, but they’re not the reason an entire genre disappeared
Rick Beato is fine but he’s entirely disconnected from contemporary guitar-based music. I agree entirely with the OP, the quality-expense ratio has never been better for this type of music.
Like with software, I'm thinking there's two different discussions: what powers the industry vs. what is possible for hobbyists.
While the industry in software is obsessed with React and K8s, hackers still like self-hosting PHP apps. Same with music. The industry is powered by highly efficient teams that write, produce, and perform music at scale for a global audience, and that's totally different from contemporary guitar-based music (I suspect!) What's possible is very different from what makes money.
Looking at her singles page on wikipedia (suspect source, but i don't know of a better one), looks like she's had 11 singles hit number 1 on billboard 100, and most of them are later songs. She's had many more in the top 10. I think your theses here is a bit suspect.
Not sure why you think my thesis is suspect. She had 12 number one singles, and they all had different writers. Those are just facts that you can easily verify.
My point is that "artists" are huge teams of people. Here are the credits for Swift's top 12 hits. Of course I wasn't in the room for the writing of any of these, but the diverstiy of the team involved is at least notable, yeah?
"We Are Never Ever Getting Back Together" (2012): Taylor Swift, Max Martin, Shellback
"Shake It Off" (2014): Taylor Swift, Max Martin, Shellback
"Blank Space" (2014): Taylor Swift, Max Martin, Shellback
"Bad Blood" (featuring Kendrick Lamar) (2015): Taylor Swift, Kendrick Lamar, Max Martin, Shellback
"Look What You Made Me Do" (2017): Taylor Swift, Jack Antonoff, Richard Fairbrass, Fred Fairbrass, Rob Manzoli
"Cardigan" (2020): Taylor Swift, Aaron Dessner
"Willow" (2020): Taylor Swift, Aaron Dessner
"All Too Well (10 Minute Version) (Taylor's Version)" (2021): Taylor Swift, Liz Rose
"Anti-Hero" (2022): Taylor Swift, Jack Antonoff
"Cruel Summer" (2023): Taylor Swift, Jack Antonoff, St. Vincent
"Is It Over Now? (Taylor's Version)" (2023): Taylor Swift, Jack Antonoff
"Fortnight" (featuring Post Malone) (2024): Taylor Swift, Post Malone, Jack Antonoff
Maybe there's been once since, but my point was she doesn't write her own songs...my point is about how the industry works now, compared to 30 years ago.
Machine-learning for audio is just a different form of audio synthesis.
That is not the issue. The issue is how incredibly generic the music is.
It also doesn't let you combine genres to make really strange sounds like audioLM can do.
This is just another Muzak generator like they use to play at Dennys. As generic music as possible to the appeal to the most average of average listener.
I think you really need to train your own model if you want to explore creative sound design or algorithmic composition. It just isn't going to be a mass market product worth venture capital money.
If you look at Taylor Swift's first 12 number one hits, each of them was written by a different writer. Compare that to bands from 30 years ago, many of whom wrote and recorded all the songs themselves.
Labels don't sign rock bands anymore because actually recording a rock band well in a studio is 10x the cost of just using a sampler and a single artist singing. I know folks want to blame AI, but it's really just enabling the latest iteration of this trend.
I'm not defending the whole thing. It's a shame, and I love going back and listening to my old Rush albums. But AI is not the problem here. It's the incentives.