AI in Music - Are We Ready (to Discuss It)?

[James Perrett]

I've just seen this paper mentioned with quite a few examples available at

https://google-research.github.io/seane ... /examples/

[BJG145]

Has anyone asked a clever (AI'ified) computer yet? surely if they intelligent then they should be part of the debate.

ChatGPT

write a poem about a bunch of musicians on a forum called "Sound On Sound" worrying that AI music will make them redundant

******************************

In Sound On Sound's virtual halls,
Musicians fret and musicians bawl,
For AI music threatens to take the stage,
And leave them all without a wage.

They wonder if their skills will fade,
And if their art will be devalued,
As algorithms and neural nets invade,
And leave them feeling quite confused.

But even as they fret and moan,
They know that music is alive,
And that the passion that they've shown,
Will always help them thrive.

So let the machines make music too,
For it can never match the soul,
Of those who've honed their craft anew,
And made it their life's goal.

And though the future may be bright,
With AI at the helm,
The musicians of Sound On Sound,
Will always have a realm.

[OneWorld]

Has anyone asked a clever (AI'ified) computer yet? surely if they intelligent then they should be part of the debate.

ChatGPT

write a poem about a bunch of musicians on a forum called "Sound On Sound" worrying that AI music will make them redundant

******************************

In Sound On Sound's virtual halls,
Musicians fret and musicians bawl,
For AI music threatens to take the stage,
And leave them all without a wage.

They wonder if their skills will fade,
And if their art will be devalued,
As algorithms and neural nets invade,
And leave them feeling quite confused.

But even as they fret and moan,
They know that music is alive,
And that the passion that they've shown,
Will always help them thrive.

So let the machines make music too,
For it can never match the soul,
Of those who've honed their craft anew,
And made it their life's goal.

And though the future may be bright,
With AI at the helm,
The musicians of Sound On Sound,
Will always have a realm.

Does a person have to be intelligent to write music and if so, how intelligent? would All Right Now be more intelligent than Gordon is a Moron?

[ManFromGlass]

Does Music Ai mean I don’t have to study scales and chords and harmony and counterpoint and arranging and modes and orchestration, and, and so I can just make music?

[Folderol]

I keep wondering who this guy named Al is.

[Drew Stephenson]

I've just seen this paper mentioned with quite a few examples available at

https://google-research.github.io/seane ... /examples/

This is a bit weird, that's my second comment that's just disappeared today.

Were you able to find anything with any more detail about how the engine works?

[James Perrett]

I've just seen this paper mentioned with quite a few examples available at

https://google-research.github.io/seane ... /examples/

This is a bit weird, that's my second comment that's just disappeared today.

Were you able to find anything with any more detail about how the engine works?

Here's the text of the paper.

https://arxiv.org/pdf/2301.11325.pdf

Sections 3 and 4 in the paper describes how it basically works although you will probably need to follow some of the references if you want the full details.

[Drew Stephenson]

Thanks James, I'll cue that up for further reading.

[Drew Stephenson]

Here's the text of the paper.

https://arxiv.org/pdf/2301.11325.pdf

Sections 3 and 4 in the paper describes how it basically works although you will probably need to follow some of the references if you want the full details.

I find this fascinating. In my head I had imagined that the AI (or machine learning I think might be more accurate in this case) would follow a similar process to a normal mix. I.e. it would synthesise a number of component parts and then add them together based on the prompts and direction. But, as I understand it, this goes straight to synthesising the complete sound. There are no component parts.
My head is struggling with this a bit.

[James Perrett]

My head is struggling with this a bit.

So is mine - I'd really like to learn more but I'm not sure where to start. By following some of the luminaries in the field I'm hoping that the odd morsel of knowledge will stick.

[ManFromGlass]

A bit too complex for me to grasp.
I assume this is all on the level but wouldn’t it be ironic if the text was generated by Ai? We no longer have guarantees about any text or images or sounds as being conceived by a sentient being.

[Drew Stephenson]

You could ask ChatGPT to create a series of prompts for and AI-based music creator...

[merlyn]

I think machine learning or neural network are better terms than artificial intelligence. This goes back to a debate between Noam Chomsky and Google's head of research Peter Norvig in the early days, where an example was "what does a cat eat?". A machine learning algorithm can answer that, but it can't tell you how it got the answer.

A human could tell you how they got the answer. "I've seen a cat eat catfood", or using a syllogism "A cat is a carnivore. Carnivores eat meat, therefore a cat eats meat." With machine learning it's statistical. The words 'cat', 'eat' and 'meat' are the most highly correlated.

There's a saying in the field "There's no data like more data." The algorithms are only as good as the training data.

[James Perrett]

A bit too complex for me to grasp.
I assume this is all on the level but wouldn’t it be ironic if the text was generated by Ai? We no longer have guarantees about any text or images or sounds as being conceived by a sentient being.

Those that are familiar with the artefacts of AI generated/enhanced audio would probably be able to tell what is real. The examples on the website I linked to are certainly not artefact free. There are those who can immediately hear Autotune and similarly, some people will be able to pick out AI audio immediately.

[Folderol]

I found most of those examples technically good, but somehow lifeless.

[awjoe]

A bit too complex for me to grasp.
I assume this is all on the level but wouldn’t it be ironic if the text was generated by Ai? We no longer have guarantees about any text or images or sounds as being conceived by a sentient being.

Those that are familiar with the artefacts of AI generated/enhanced audio would probably be able to tell what is real. The examples on the website I linked to are certainly not artefact free. There are those who can immediately hear Autotune and similarly, some people will be able to pick out AI audio immediately.

A buddy of mine sent me a series of Chatgpt pieces, each slightly better than the previous one. The reason they got better was that his input, the core idea, for each one was more imaginative than the previous one. He was in awe of its speed and ability to genre-shift. I was disappointed by the lack of creativity of expression. It was about as close to the real thing as the self-checkout voice at my supermaket saying: 'Thank you for shopping at Thrifty's' as it spits out my receipt. Chatgpt doesn't do creative, it does generic. No matter how big the database it has to draw on, Toonzgpt is always going to be closer to Kenny G than John Coltrane, and not even that.

[merlyn]

The 'P' in GPT stands for 'pre-trained'. And the training data is ... the entire internet.

[ManFromGlass]

I wonder if that means too much choice as sometimes it’s what you leave out that brings the magic

[awjoe]

I wonder if that means too much choice as sometimes it’s what you leave out that brings the magic

I bet you could program that.

[merlyn]

I wonder if that means too much choice as sometimes it’s what you leave out that brings the magic

GPT-3 is a lot better than GPT-2, and it's ten times the size. More data doesn't mean the algorithm can pick and choose. Once it's trained the algorithm kind of is the data (perhaps more correctly the relationships contained in the data). OpenAI, who came up with GPT, seem to think transformers (the 'T') are scalable, and GPT-4 will be even bigger, and maybe write better poetry.

MusicLM ... conceptually I'd say it's there. You would think that would scale too.

[Martin Walker]

A buddy of mine sent me a series of Chatgpt pieces, each slightly better than the previous one. The reason they got better was that his input, the core idea, for each one was more imaginative than the previous one.

In a way, this is similar to what happens when we buy a new instrument and explore its sound design options. Our first presets (and indeed the factory sounds) may be boring, but as we explore the options in more detail we end up being more creative.

If AI encourages us to be be more imaginative, then so be it - this could be a benefit in disguise, but bypassing much of the slog by letting us see the results in double-quick time!

[merlyn]

... All we have to do is relinquish our hold on the top position of the hierarchy on this planet and continue to build and service the machines. (The machine overlord idea isn't that drastic, really, considering how close we are to it already. Some people would even argue that we're making a hash out of our stewardship of the planet anyway, so maybe it's time we stepped down.)

Going by your more recent posts you may have changed your mind, but this is such a bad idea it's difficult to believe. Part of domesticated primate's (that's us) lack of stewardship abilities is that we burn fossil fuels to generate electricity to power data centres to run neural networks that chat, and write average poetry.

[awjoe]

It was tongue in cheek.

[BJG145]

This is a bit random, but it's kind of on topic. It's something I just posted in the forum for the Artiphon Orba.

The background is that this is a cool device with terrible support. Artiphon have written some dreadful software for it and refused to document or explain it. The community has been focussed on reverse-engineering it to see how they can improve it. This post summarises recent experiences with ChatGPT and programming.

* * * * * * * * * * * * * * * * * * * * * * * * * * * *

ChatGPT v Orba 1

Part 8

ChatGPT and its ilk tend to get a bit of a bad press.Most of the news articles now are about how soon it's expected to cause human extinction. You're also on thin ice when you start discussing programming. I'm aware that there are several talented programmers on this forum who I highly respect and wouldn't touch it with a bargepole. When people started using ChatGPT to try and generate code for Monome products and posted it on Lines (Monome community forum), they didn't get a warm welcome...

...nevertheless, I'm a fan. It's the arrival of ChatGPT that prompted me to re-engage with the Orba, and with programming in general, and I've found this quite rewarding.

To some extent you can engage with ChatGPT directly on this stuff, but it's very wayward and unpredictable, as you'll know if you've spent any time with it. It has flashes of brilliance but never does anything the same way twice. It's forgetful and makes basic errors. That's just its nature.

I've spent a number of hours now engaging with ChatGPT in discussions about note sequence data, and found the best approach for me was to use it to write Python routines. I like the idea of using ChatGPT as a human-language-level programming tool ("...create the Orba XML file to play Beethoven's Fifth on an ocarina...") but it's too frustrating to try and work that way, partly because it's so forgetful. It will give you the right answer one minute and the wrong answer the next. So I started working with it via Python instead. Because at least the action of a Python program, once it's been established, is predictable. It takes quite a lot of effort to get it to produce working code, and I expect it's not good code, but for a non-programmer it can nevertheless be a useful crutch. With patience.

I'll present the code at some point, after I've refined it. I've been developing routines to convert between Base64 eventData strings, raw "note block" data, formatted human-readable "note block data"; routines to modify note data with new timing information, and interpret note data to produce "console output" with lists of notes and event times.

I was pleased with today's progress, and finally had a set of Python utilities that allowed me to create the XML to play a two-octave ascending and descending chromatic scale on Orba 1. Without ChatGPT, I would have found that so difficult that I wouldn't even have tried.

The challenge I've set myself is to sequence the opening of "Flight of the Bumble Bee". Unfortunately ChatGPT doesn't seem familiar with the note data for it so I'll have to round it up from somewhere else. If I can get that far I'll post up a YT link and some code.

[DC-Choppah]

When you call a company and realize you are talking to a computer you feel offended. When you chat online for help in a medical situation, you are offended by talking to a computer. The AI has no judgement and can only make you more annoyed. Where is the real person? Why are you making me talk to this stupid robot? You must hate me? Humans have no patience for AI since it has zero intelligence.

AI music is offensive to most people and is just a fad. When the market looses interest in it and even punishes it, since it is so offensive things will change very quickly. Money talks. AI music is like spam.

Real concert halls with real people are full. AI is not welcome there.

AI is just software automation and is just a tool. If you find uses for it in making music to relegate the tedious stuff to a machine that's great. That is what the machines are for.