This May Be Generative AI Music's Killer App: Extend Your Clips

[dubbmann]

Hi all

The AI rev continues apace and I thought this milestone was so huge it merited an update here. The latest feature of Udio and Suno allows you to upload a piece of original music. You specify what you'd like it sound like (that's your 'prompt') and hit enter. The AI will generate multiple possible takes. If you specify vocals it will add a vocalist. You can submit your own lyrics or ask it to make lyrics based on your suggested topics.

Here's a non-AI musician giving his reactions and showing a dozen or so examples of snippets he's uploaded and what Udio sends back.

https://www.youtube.com/watch?v=INkCR1uGrSU&t=545s

This to me is huge b/c its not asking the AI to create anonymous library music (which it's accomplished) but rather it asks the AI to take your tunes, learn, and add. It's like a musical amanuensis. I'll bet Eno already has it up and running in his studio and home

[sonics]

This to me is huge b/c its not asking the AI to create anonymous library music (which it's accomplished) but rather it asks the AI to take your tunes, learn, and add.

But the generator is using its stolen library to add to your ideas, so as soon as anything is added, it's no longer really your music.

You would have to train it on your music alone for any output to be "authentic".

[muzines]

But the generator is using its stolen library to add to your ideas, so as soon as anything is added, it's no longer really your music.

I mean, that's pretty much the way humans work. You train your own brain on thousands of hours of music you listen to, absorb the things you like, and what you don't like, and those things because part of your neural processing network that you use when creating new pieces. And you of course find your own things that also go into that evaluative model as well, whether by accident or design.

It's not like you're consciously directly lifting melodies or chord sequences from other things (although we do that sometimes too), it's more than those things all go to influence and shape the kinds of things we come up with - that's pretty much what's happening with these AI tools too. It's not like it's recording existing melodies, and just cutting/pasting bits from other people's work - it's tuning the model, effectively developing it's own likes (more similar stuff is regarded as more successful, less similarity means those ideas are less successful, and thus weighted less in the output.

[dubbmann]

Here's another YT vid w a muso reaction

https://www.youtube.com/watch?v=KZRcZ4xBM2A&t=327s

But yes, the legality/illegality issue is definitely live. But imagine an AI trained on classical music that's past copyright. Or use music played from MIDI sources, again no infringing. Eventually the labels will license it and it'll raise the prices which is fair. But this new feature to me is what I've been waiting for. I don't anonymous library music. I want my anonymous library music!

[Dynamic Mike]

Hi all

The AI rev continues apace and I thought this milestone was so huge it merited an update here. The latest feature of Udio and Suno allows you to upload a piece of original music. You specify what you'd like it sound like (that's your 'prompt') and hit enter. The AI will generate multiple possible takes. If you specify vocals it will add a vocalist. You can submit your own lyrics or ask it to make lyrics based on your suggested topics.

Here's a non-AI musician giving his reactions and showing a dozen or so examples of snippets he's uploaded and what Udio sends back.

https://www.youtube.com/watch?v=INkCR1uGrSU&t=545s

This to me is huge b/c its not asking the AI to create anonymous library music (which it's accomplished) but rather it asks the AI to take your tunes, learn, and add. It's like a musical amanuensis. I'll bet Eno already has it up and running in his studio and home

The link on your signatures comes up as domain for sale. You may need to split them

[Folderol]

Definitely not coming here!
No way would I regard (nor promote) something like this as my music

[BJG145]

I briefly signed up for Udio a while back to try this; hadn't noticed the Suno version. But yes, this is far more interesting than randomly generated complete tracks, and a step towards AI music generators becoming useful tools for musicians.

[OneWorld]

Well at the side of this, makes even my music sound reasonable.Nope still sounds bit manufactured. I guess if you're selling library music to play in the background at a £1 shop then ok this might pass muster but Taylors Whiffed doesn't need to sell her private jet yet.

I heard tell that Microsoft are going to start charging a monthly subscription for ChatGPT even though a recent survey showed it prevailed by 60% to 40% during a recent set of Turing Test experiments, but who prepared the script? Those that make such fabulous claims for it no doubt, but how would it go on dealing ambiguous metaphor in unseen text?

Hats off to the writers of the code but instead of fish fingers, I'd still prefer to stick with a Fish & Chip supper

[dubbmann]

One interesting use: have it generate middle 8s for your verse/chorus upload. Lots of folks can't write good bridges.

[sc1460]

I think you’re right, it is huge, maybe not right now, but 3-5-7 years the progress is going to be totally shocking.

It is based on plagiarism, but that’s not new is it? It’s just on a mass scale undreamt of before. Worse as it’s based on probability it kills the minority tastes that drive creative inspiration. But it’s great for firms churning out mass produced art.

Today we can trawl through midi libraries, sample banks and it takes a lot of time. With our $20 a month AI music assistant it’s quicker and easier and we can steal through musical history.

I just told a bassist, they wanted James Jameson, why give me a Bootsy Collins riff? Bassist says he’ll need 2 weeks to change as he’s gigging. Singer can’t wait two weeks, don’t have budget for new bassist etc etc may as well replace the bassist with AI, no? Didnt Berry Gordy and StockAW and countless other producers replace musicians with others? Except now….

This is no doubt a descent into a musical hell of blandness. 600 share-millionaires in the USA last year alone investing in AI firms. Indeed the first AI generated movie is now out. Enjoy the ride, invest in the bubble.

Cheers
H

Hi all

The AI rev continues apace and I thought this milestone was so huge it merited an update here. The latest feature of Udio and Suno allows you to upload a piece of original music. You specify what you'd like it sound like (that's your 'prompt') and hit enter. The AI will generate multiple possible takes. If you specify vocals it will add a vocalist. You can submit your own lyrics or ask it to make lyrics based on your suggested topics.

Here's a non-AI musician giving his reactions and showing a dozen or so examples of snippets he's uploaded and what Udio sends back.

https://www.youtube.com/watch?v=INkCR1uGrSU&t=545s

This to me is huge b/c its not asking the AI to create anonymous library music (which it's accomplished) but rather it asks the AI to take your tunes, learn, and add. It's like a musical amanuensis. I'll bet Eno already has it up and running in his studio and home

[RichardT]

This could be a lot of fun. Sometimes it would be great to generate a string arrangement automatically without putting the work in.

Especially if I could say ‘do it in the style of Ravel’. I love his harmonies and I don’t understand them at all.

[merlyn]

But the generator is using its stolen library to add to your ideas, so as soon as anything is added, it's no longer really your music.

I mean, that's pretty much the way humans work. ...

A neural network isn't a human, though. It's a machine. If someone infringed copyright using a photocopier (a machine) to photocopy a book, it's not a defense to say that "Yeah, but a human could copy it out longhand."

[w oxo cube]

Doesn't sound like a great deal of fun, the enjoyment that I get personally from writing music is the playing and generation of ideas that are soundtracking my own thoughts and emotions. I can't imagine being in the zone and thinking, "ah let's break off and feed prompts into a computer and let it finish it for me."

[Forum Admin]

Where I see AI being of use to musicians is in idea generation or extrapolation, especially with lyrics.

I never found it hard to write instrumental music, but lyric writing was pretty hard for me - never came easy - so any tool (AI app) that could assist here would get a look in, if it was me using it.

I wouldn't let it take away all the fun bits of creativity, but if I hit a roadblock musically, and wanted it to suggest some lyrics that could be used as verse 3 and 4, say, based on what I'd written for verse 1 and 2 and a chorus, then I'd probably try using the app to help me.

Just as DAWs have incorporated Chord generators and Lyric playback tools, I can very much see them adopting ChatGPT-style prompts inside the DAW to aid with songwriting. Nobody will be forcing anyone to use such tools, but having them integrated would be handy.

[RichardT]

Where I see AI being of use to musicians is in idea generation or extrapolation, especially with lyrics.

I never found it hard to write instrumental music, but lyric writing was pretty hard for me - never came easy - so any tool (AI app) that could assist here would get a look in, if it was me using it.

I wouldn't let it take away all the fun bits of creativity, but if I hit a roadblock musically, and wanted it to suggest some lyrics that could be used as verse 3 and 4, say, based on what I'd written for verse 1 and 2 and a chorus, then I'd probably try using the app to help me.

Just as DAWs have incorporated Chord generators and Lyric playback tools, I can very much see them adopting ChatGPT-style prompts inside the DAW to aid with songwriting. Nobody will be forcing anyone to use such tools, but having them integrated would be handy.

Yes, that would be good. At the moment lyrics and poems generated by AI are far worse (imo) than the music. They are absolutely dreadful! But I’m sure that will change.

[OneWorld]

One interesting use: have it generate middle 8s for your verse/chorus upload. Lots of folks can't write good bridges.

Actually you could have a point there. I wrote a tune for a friend of mine, as a birthday present, and she was really impressed with it, she liked it.

Then as an extra I thought I'll put through AI and send that to her, joking a singer friend of mine had done her version of it - of course it was a full AI treatment, I just uploaded the AI sans vocals but with the lyrics, and I was quite impressed, anyway I send it off to my friend, she couldn't tell the singer was a virtual singer, and more importantly went to say "WOW, it is so catchy" Fo the bridge I tried to avoid cliched harmony, but when I tend to do that, I over egg the pudding and it sounds more like Schoenberg, but the AI did a lovely job -- where the AI singer did struggle was the phrasing but in fact it sort of added to the quirkiness.

And that reminded me, a new technology night not neccessarily exist to displace existing technology but come at it with fresh ears so to speak, compliment it, not replace it.

The jet engine didn't see the extinction of the turbo-prop.

[sc1460]

On the news feed:
“Earlier today, the RIAA announced that it had filed a lawsuit against AI music generation platforms Udio and Suno, with the plaintiffs including Sony Entertainment Music, Warner Records, and Universal Music Group. The lawsuit seeks damages of up to $150,00 for each piece of infringed work, and to stop the two AI companies from training on the label’s copyrighted songs.”

Given it takes a HUGE amount of stolen music to train a LLM, that could be huge $$$$ amount

Soon it will be AI lawyers suing other Ai lawyers…. musicians will seek the advice of AI lawyers to sue producers and labels, who will have their own AI lawyers….this is just awesome.
Cheers
H

[Drew Stephenson]

Given it takes a HUGE amount of stolen music to train a LLM, that could be huge $$$$ amount

Except I'm not aware of any court that's yet decided that listening is the same as stealing.

[BJG145]

Training an AI on music isn't the same as listening to it though. Does the method involve making a copy...? Dunno.

When you upload a clip to Udio you have to confirm that you have the copyright and distribution rights. I wonder what it is about this process that differs from the methods used by its creators.

I'm not saying it's right or wrong, and as discussed previously, things may work out so that it's no different from if a person did the same thing...it'll be judged on its similarities rather than the method of creation. But perhaps the method of creation in itself is still a grey area.

[Folderol]

Given it takes a HUGE amount of stolen music to train a LLM, that could be huge $$$$ amount

Except I'm not aware of any court that's yet decided that listening is the same as stealing.

It doesn't need to. The law is concerned with mechanical copyright. An LLM is a mechanism, human ears are not.

[Drew Stephenson]

But it is not making or taking a copy. The track is listened to and assessed, this adds data points to the neural network but it is not taking a copy. There is no more copyright infringement than if you or I listen to a track and learn it.

The mechanical bit is about reproduction licences, nothing to do with listening.

People need to be very careful about the future they're wishing for. Any law that is created can and will be abused. The worse the design and implementation of that law, the more scope their is for abuse by those who are in a position of power.
If the RIAA are successful in persuading a judge that listening requires a licence, do you think they'll stop with machine listening?
Any study of their behaviour (or similar groups like the BPI, ASCAP, GEMA, MPAA etc.) over the last 50 years would tell you that this would just be the first slip down a long slope.

[BJG145]

I’m not convinced. For one thing, “listening” is a human activity, not an AI one, so it’s a misleading term. Is laying down these data points different from laying down magnetic points, or optical points, as far as the law is concerned…? Are we sure that no part of the process of ingesting these tracks into an LLM is copying of any sort? And if not, why can people only upload their own material, while the company can upload whatever they like? There’s a number of unanswered questions and shady procedures here.

A lot of creatives are uncomfortable with some company feeding their life’s work into a computer and churning out facsimiles for profit with no acknowledgement, and I can understand those concerns. Some more transparency would be a good start.

[Forum Admin]

If the RIAA are successful in persuading a judge that listening requires a licence, do you think they'll stop with machine listening?

It's already happened years ago.

In the UK, if you wish to listen to music in a hairdressers, or hotel foyer, or office building over the tannoy, etc you legally have to buy a PRS license. We used to have one at SOS HQ back in the days when we had a physical office we all attended.

[Drew Stephenson]

I’m not convinced. For one thing, “listening” is a human activity, not an AI one, so it’s a misleading term. Is laying down these data points different from laying down magnetic points, or optical points, as far as the law is concerned…? Are we sure that no part of the process of ingesting these tracks into an LLM is copying of any sort?

You have to look at what is actually copyrightable. The recording and the composition. There is no copyright infringement on taking a load of data points that analyse that music in the same way that there's no copyright infringement in my writing a load of notes about my analysis of a piece of music.

And if not, why can people only upload their own material, while the company can upload whatever they like?

Probably because the companies know that if they allowed people to load up copyrighted music then they would (because most people don't understand copyright) and then there'd be a very real risk of them retaining, for however short a period of time, a copyrighted file.
Also to try and avoid court cases like this.

A lot of creatives are uncomfortable with some company feeding their life’s work into a computer and churning out facsimiles for profit with no acknowledgement, and I can understand those concerns. Some more transparency would be a good start.

Absolutely, I too share some of the concerns. But copyright is a hammer and this is not a nail.

It's already happened years ago.

In the UK, if you wish to listen to music in a hairdressers, or hotel foyer, or office building over the tannoy, etc you legally have to buy a PRS license. We used to have one at SOS HQ back in the days when we had a physical office we all attended.

Not strictly true Ian, the licence is required to broadcast the music not listen to it. Hence the case of of the woman being told she needed a licence to play music to her horses.

[Folderol]

If the RIAA are successful in persuading a judge that listening requires a licence, do you think they'll stop with machine listening?

It's already happened years ago.

In the UK, if you wish to listen to music in a hairdressers, or hotel foyer, or office building over the tannoy, etc you legally have to buy a PRS license. We used to have one at SOS HQ back in the days when we had a physical office we all attended.

That doesn't sound right.
Surely it is the venue that needs the licenses not individual visitors. So, I think it's reasonable to suggest that an LLM is a venue rather than a 'listener'.