• Hi and welcome to the Studio One User Forum!

    Please note that this is an independent, user-driven forum and is not endorsed by, affiliated with, or maintained by PreSonus. Learn more in the Welcome thread!

AI Insanity

Now, if AI could analyze a comped vocal and immediately toss out the takes that aren't as good, then I would definitely be a fan!
That would be an amazing tool to add! Maybe we can ask some AI coding machine to write it?
 
... Now, if AI could analyze a comped vocal and immediately toss out the takes that aren't as good, then I would definitely be a fan! ...
Interesting, as that would be the exact opposite of how I'd use it, leaving the decision on what's better with the machine.

I'd rather see AI as a 24/7 volunteer in the studio I can tell to add some violins after verse 2 in the style of *Example A*, *B* and *C*. And then I pick the one I like best, or have it do 3 more with additional instructions.
 
Just a thought but, interestingly when actual intelligence meets artificial intelligence do you get something like this...

To view this content we will need your consent to set third party cookies.
For more detailed information, see our cookies page.

Regards
Indeed you can. With Bowie, an innovator, who started as a poet, and with pantomime, almost any stimulus will turn into something quite imaginative. Many of us will also use it as a stimulous for an idea, a group of lyrics that might work and also be imaginative. As for the general usage among many, people will find ways in which to make AI gum-ball machine usage work for their art form (or lack thereof). Some will mock it's cold empty shell.
Thanks for sharing the vid, btw.
 
Last edited:
Interesting, as that would be the exact opposite of how I'd use it, leaving the decision on what's better with the machine.

But that's what's potentially cool about AI, although in this case, there's also machine learning involved. This kind of process builds up a database over time. So ultimately, anything it did would be based on decisions I had made in the past. For example, it would know that I like vocal takes where there's vibrato on sustained notes at the end of a phrase, or that one take had pitch problems but another one didn't, or that sibilance would be difficult to fix. It could also flag a take if its phrasing wasn't used in any other take...that sort of thing. Then, it would give me candidates that matched my criteria for good takes, and I'd choose the one I wanted. I wouldn't have to spend time listening to takes I was likely going to reject anyway.
 
Would AI be able to pick the 'most compelling' takes then? Many if not all issues you mentioned can be fixed. Melodyne has the tools and no doubt AI will be able to use those (eventually), automatically applying the edits per your personal preferences. That would make all takes pass your aforementioned criteria, leaving you where you are now. 'Most compelling' on the other hand is much more elusive but for me also most decisive for selection. Not criticising, just explaining why I'll prefer to make the (pre)selection myself rather than let AI do it :)
 
Last edited:
  • Like
Reactions: AAV
Scary good as in if I hadn't broadcast that it was completely done by AI, nobody here would have even known that.

That's all I meant by 'scary good'. (y)
In all honesty, I have to say I was disappointed on the realism end, from your example. That's what prompted my initial response. Musically, it's not very convincing at all. On a positive note, I did think the lyrical content could possibly hold up if given the right set of circumstances.
Vocals were an extremely poor facsimile, and noticed from the first syllables, to the last.
We have very different opinions on if AI sounds convincing or not.
 
Last edited:
Not criticising, just explaining why I'll prefer to make the (pre)selection myself rather than let AI do it

Yes, but it seems I'm not clear about what I want. My goal would not be for AI to edit or choose the comps I would use. The goal would be for AI to isolate only the comps that had desirable characteristics in common with comps I'd chosen in the past, thereby getting rid of the comps I wouldn't pick. It could very easily present several comps of the same section as candidates if they fit my criteria, and then I could choose the one I wanted.

Currently, I have to listen to every comp to find the good parts. After finding all the good parts, then I compare them in context with the music using my Dim Solo trick. The technically "best" comp is not always the artistically "best" comp, so I choose the one I like best.

I'd love it if AI could save me the time of having to listen to everything just to find the good stuff. I usually don't do more than 4 or 5 comps because of the hassle in auditioning more comps than that. It would be great if I could do 10 or 20 comps, but end up with only 3 or 4 candidates per comp I had to compare, instead of 10 or 20.
 
The music AI was trained on 100% copyrighted material, with the purpose of generating revenue by "capturing" revenue from the people who created the material used for training.

The AI companies take the position that copyright doesn't apply because the use in training was never defined, therefore is legal. My take is that if a use has not been defined, then those rights belong to the creator. Not to the entity that may have bought the rights previously - they paid for the uses that were defined at the time of sale. But obviously, the world doesn't agree with me...

There are places where AI has great potential and great value. It can be argued that AI will not be able to create truly great art. But artists aren't born great - they work their tails off, producing a lot of not great art on the path to getting good. If AI can outdo them through most of their journey, will they bother?
 
The AI companies take the position that copyright doesn't apply because the use in training was never defined, therefore is legal.

They also claim fair use, because it's transformative and it doesn't use large sections of existing art. I'll give them that. However, one of the tests for fair use is whether it negatively impacts the market for the original work or its potential earnings. If the use replaces the need for the original, it’s less likely to be considered fair use. But which part is the "original" part that it's replacing? Or is it more like "class action" replacement? Another test is the intended use. Using something with the intent to educate is more likely to be considered fair use than if the source material is used for commercial purposes. As I wrote in a Mixonline column a couple years ago, maybe AI stands for "Attorneys Incoming" :)
 
The point of "it doesn't use large sections of existing art" I won't concede. There have been cases where a prompt has generated large portions of an existing work, and indeed there are lawsuits pending (as you said "Attorneys Incoming") for some of these. And if you're using small portions of many existing works to create a new work, well, law has already determined that compensation is required.

So, if they are claiming fair use, then have them show "fair use of exactly what". That would be interesting.

In any case, I am just the "man shaking fist at clouds". With all the money behind AI, and on the table for the AI victors, I don't see that the creatives will be treated well.
 
After experimenting there with most of the genres, I've come to one obvious conclusion, modern Hip-Hop / Rap is no danger from AI. at least not on that website. :) Maybe somewhere else where the coders understand the genre better.

While it's easy enough for AI to generate a basic hip hop or rap beat with a thumping kick drums and all that, and some rhymes, I could never get it to create someting remotely muscially interesting, but that's not the worst part. Putting the musical beat aside, those genres have evolved quite dramatically in regards to lyrical timing and flow, and AI doesn't seem capable of mimicing that yet. The lyrical flows are kinda stuck in the past, cornball timing from the Run DMC era :)

In context, it does a decent job with Rock, Pop, R&B and Country. Some of those result were not bad at all. Dance and EDM on the other hand, kinda sucked. Metal genre was musically ok.

Anyway, it was fun. Given that 99% of the planet are not musicians and songwriters, it serves a definite purpose in the market. If a random person wants to give someone an original love song for their birthday or whatever else, it's easy and not expensive.
 
Last edited:
The music AI was trained on 100% copyrighted material, with the purpose of generating revenue by "capturing" revenue from the people who created the material used for training.

The AI companies take the position that copyright doesn't apply because the use in training was never defined, therefore is legal. My take is that if a use has not been defined, then those rights belong to the creator. Not to the entity that may have bought the rights previously - they paid for the uses that were defined at the time of sale. But obviously, the world doesn't agree with me...
Copyright protection has been defined. Of course that differs depending on the country, but nonetheless, there are guidelines to a release. Any release. AI is compiling any data out there, and as we know, that can be had with anything from legitimate purchases, to Napster and then some (and not be caught). So if an AI developer even came up with the lyrics "Sweet dreams are made of these", there wouldn't be copyright infringement. Six words in succession are acceptable. Now if that were followed by ....Who am I", then Eurythmics would have a legitimate case, and pursue that. The same as musical content has a limit. So this is really no different than if any artist sat and listened, recorded, or even claimed they guessed some existing copyright release. We can capture anything we want (although don't get caught for stealing it). The act of steeling an artists work is in the release, and claim of originaly, or performance. Also staying within the alloted limits.
AI simply has a larger shopping cart going down the musical aisles. That's not cause for any lawsuit I'm aware of. It's in the release of one's work, where the lines in the sand are drawn. Even for AI (ownership) to abide by.

Training is a different form of legal premise used now for largely playing someone else's music on YouTube, or the web in general. So I may be missing something to your point. I'm just bringing up their fair use. If they were flooding the airwaves with nothing but their ...uhem... music, then that's likely another legal case
There are places where AI has great potential and great value. It can be argued that AI will not be able to create truly great art.
Art, is an illusion.

Just a brief and fun reference here, you may enjoy. It's from the Moody Blues. Written by drummer Graeme Edge (last part of "Late Lament").
"Cold-hearted orb that rules the night
Removes the colours from our sight
Red is grey is yellow white
But we decide which is right
And which is an illusion".


I might be wrong, but I don't think AI could pull off that kind of poetic wonder. 😉
 
Last edited:
The point of "it doesn't use large sections of existing art" I won't concede.
In the context of fair use, that relates to sections of a specific piece of art, i.e., something you recognize as being like something else. For example, George Harrison's "My Sweet Lord" was deemed to infringe because sections of it were deemed virtually identical to large sections of "He's So Fine."
 
Yep, and the Chiffons were awarded over half a million dollars for Harrison's intended or not, plagiarism. The lines drawn in the sand.

I wonder how that breaks down to some of the more common melodies such as standard blues progressions. I guess there are allowances. In any event, AI owners I suspect would have to play by the same rules. I wonder if I can still contact that entertainment lawyer I knew many years ago. 🤔 Anyway, my three cents, and thanks Craig and others for the numerous responses to this intriguing subject.
 
When it comes to AI, the legal and societal consequences are being made up as we go along. Being musicians, we of course understand improvisation :ROFLMAO:
 
If a random person wants to give someone an original love song for their birthday or whatever else, it's easy and not expensive.

Brilliant, I hadn't considered that.

Hey - since you came up with the idea, why don't you start an AI-based "Audio Greeting Card" company and become a millionaire? People fill out a form with information that allows you to create an appropriate prompt, you enter it into your AI Music Generator, push a button and voila! They pay you $20, and they have their custom audio greeting card.
 
Brilliant, I hadn't considered that.

Hey [@ Gray Wolf] - since you came up with the idea, why don't you start an AI-based "Audio Greeting Card" company.....
I just googled that, and there's already a slew of them. But hey, one more couldn't hurt. 😉
 
When it comes to AI, the legal and societal consequences are being made up as we go along. Being musicians, we of course understand improvisation :ROFLMAO:
intellectual property.
🥸
Groucho Marx: It's all right, that's in every contract. That's what they call a sanity clause. Chico Marx: You can't fool me, there ain't no sanity clause.
"Attorneys Income_ing" , there have been lots of legal battles over the years and even between band members.

I'm quite happy having time and space to be on my own creative journey and to have the odd jam with a few friends. I imagine there are lot's of folk that enjoy just being creative.

Best regards to all.
 
  • Like
Reactions: AAV
Meanwhile I still get mixed results out of programming AI. I'm just glad it is C++ right now because the type system catches many things at compile time. In Python you'd be screwed.
 
There are some very useful uses for AI. One being the ability to transform a vocal scat into say a Middle Eastern plucked instrument playing the exact same melody. This is cool. But it requires you to come up with the decent melody to start with. AI is not composing now in this sense, it is transforming. The thing is, does an AI created production actually really move you emotionally. You will find in many cases it misses out in this area. Something that real writers and composers can do.
 
Back
Top