Did AI just kill classical music?

[Alan Howe]

<<Violinist and creator Daniel Kurganov has released a new video. Listen to it here and see if you start to ask yourself some questions...
https://www.thestrad.com/video/did-ai-just-kill-classical-music-a-video-from-violinist-daniel-kurganov/18010.article

Kurganov shares: 'How do we understand what is good string playing? What exactly have we learnt about the intricacies of nuance and beauty? Is it a mystery? Can it be taught or understood in explicit terms? Or can it only be emulated and passed on from one emotional being to the next? I would like to gently challenge all of our preconceptions with this sample of music.

'What you are hearing is music entirely composed by and played by Artificial Intelligence. My role was minimal, limited to guiding the AI with simple text instructions. This represents a groundbreaking moment advancement in technology—and this is the dumbest these tools will ever be. So, we must ask: are we seeing the end of classical music as we know it, or does this mark a new beginning?'>>

[John Boyer]

Interesting topic, Alan, but does it really pertain in any way to our remit, unsung composers?

[Alan Howe]

I take your point.

I just wondered whether AI-derived classical music (which is surely going to be unfamiliar) counts as 'unsung'? In the above brief article the violinist Daniel Kurganov says that he had some sort of hand in 'creating' the new music in the video...

[der79sebas]

I ask myself if AI will soon be able to create good recordings out of (possibly handwritten) scores. This would be a great possibility to listen to unsung music which would otherwise never be performed or recorded (e.g. the much too big operas of Wayditch).

[Alan Howe]

That's a very exciting prospect, if it's a possibility.

[TerraEpon]

I remember there was some experimenting with computer composed classical-esque music in the early 90s. This really isn't a new thing.

[tuatara442442]

It has no sense of form, theme and development, so it can't generate real counterpoint. I think the mechanism of large language model dictates it, that is predicting a likely next step using a given last step. While I think there is certainly the possiblility of it referencing some further past material that it processes, it doesn't know abstraction and "understand" what macroscopic pattern is. This, I think, is exemplified by its inability to correctly produce images of intricate decorative patterns.
The way to solve it, I think, is to let another AI trained with pattern abstractions edit the "source material". (Or human arrangers, which results in the rise of their importance).
There are two ways of AI composing, one is audio generation, the other is midi file generation. It is hard to edit the music itself with only an audio file and is defeating its own purpose transcribing it to sheet music. But an AI arranger of midi file could be trained. And I think an arranger AI could be trained on abstracted polyphonic inner workings, and it will rely on midi renditions of existing pieces as its training material. So it could be a good news for midi rendition makers. 

[tuatara442442]

I remember there was some experimenting with computer composed classical-esque music in the early 90s. This really isn't a new thing.

Yeah, David Cope is a big name among them. Centaur published recordings of his computer-composed music, as well as 39 volumes of Consortium to Distribute Computer Music (CDCM) recordings with compositions produced by many people. That was many years ago

[tpaloj]

I ask myself if AI will soon be able to create good recordings out of (possibly handwritten) scores. This would be a great possibility to listen to unsung music which would otherwise never be performed or recorded (e.g. the much too big operas of Wayditch).

I often ask the same and have been casually following improvements in OMR (optical music recognition) software. It appears to me that handwritten recognition is still far off, since these programs are still cping with some basic issues in recognizing printed music. But I think we're getting close to submitting any printed notation to such software and having it produce a reliable, decent quality audio rendition without too much user input at all.

I think mainstream pop/rock music makers are much more in danger of losing their jobs to AI tools than the classical scene, where the market is (at least still for the most part) rather more conscious of the performers' actually performing music to an historically rooted and informed standard.

[Alan Howe]

Thanks. That's a useful update on current developments.

[tuatara442442]

But I think we're getting close to submitting any printed notation to such software and having it produce a reliable, decent quality audio rendition without too much user input at all.

If you are saying generating reality-quality other than midi-quality performance, then there won't be miserable circumstances like Ulrich Leykam "conducting" literally his own "Bayreuth Digital Orchestra" to accompany Naxos-Marco Polo's S. Wagner Sonnenflammen production.

[tpaloj]

But I think we're getting close to submitting any printed notation to such software and having it produce a reliable, decent quality audio rendition without too much user input at all.

If you are saying generating reality-quality other than midi-quality performance, then there won't be miserable circumstances like Ulrich Leykam "conducting" literally his own "Bayreuth Digital Orchestra" to accompany Naxos-Marco Polo's S. Wagner Sonnenflammen production.

Ooff. I listened to some samples. It sounds like it would have been way more straightforward to hire an actual orchestra given the technical work involved. But I don't know what circumstances led to them performing the work in this arrangement.

It should be added that while orchestral music can be made to sound somewhat realistic via technology with some effort these days, not even AI can seem to fake good vocals, never mind operaic singing voices to any believable level yet.

[der79sebas]

The synthesizer "Sonnenflammen" were done during Corona and this production was some sort of protest against cultural lockdown. Well, it was worth the try, but of course this never should have been published as DVD...

[Ilja]

But I think we're getting close to submitting any printed notation to such software and having it produce a reliable, decent quality audio rendition without too much user input at all.

If you are saying generating reality-quality other than midi-quality performance, then there won't be miserable circumstances like Ulrich Leykam "conducting" literally his own "Bayreuth Digital Orchestra" to accompany Naxos-Marco Polo's S. Wagner Sonnenflammen production.

Ooff. I listened to some samples. It sounds like it would have been way more straightforward to hire an actual orchestra given the technical work involved. But I don't know what circumstances led to them performing the work in this arrangement.

It should be added that while orchestral music can be made to sound somewhat realistic via technology with some effort these days, not even AI can seem to fake good vocals, never mind operaic singing voices to any believable level yet.

There are various generative AI engines today that have no issue producing somewhat convincing vocals, or instrumental music for that matter. It's just that "classical" music has not been the focal point for the best efforts sofar; rather, those have been directed at other, more popular genres. As an example, I just asked for a "gothic song about bananas" on Suno and got this - frighteningly realistic. Also witnessing AI developments in my own field I am convinced that, with time (and money), this technology could make great strides very quickly.

[Alan Howe]

I've always wondered a mature Wagner symphony might sound like. Any offers from you AI whizz-kids out there? 

Oh, sorry. Scriabin wrote three in that idiom.