I've spent the past 24 hours being quite astonished at Udio, which I already consider vastly superior to any AI music generator I have heard to date.
To be clear: I share this for the purposes of awareness and conversation, not endorsement.
The lyrics I gave it were random text relating to a conversation I was having with a friend this morning. Since Udio only gives you 33 second clips, I had to extend it three times (adding an instrumental into, 2nd set of text and final, largely instrumental outro)
BTW, as of this morning, Udio is claiming it is removes artist names (here "Richard Strauss") from the prompts, but as you can clearly hear (if you know his music), they must still be using it in the underlying mechanics
---
Good Morning Soo
The dogs need a walk
Can you hear them?
I hope you have a lovely day
Are you getting the groceries and the meal kits?
Yes, this day is for shopping.
How marvelous!
---
You need to be a member of composersforum to add comments!
Replies
I don't really share Dave's pessimism about AI.If AI can produce the commercial stuff that humans currently do then it will inevitably happen because it's cheaper. And someone still has to programme the AI so I expect we'll be more in a world of hybrids where composers use AI to save time rather than being completely put out to grass although that may happen in some cases. And yes, as Wesley says, it's all but inevitable that some composers will pass of AI stuff as their own work.
I still think we're some distance away from AI taking over concert music though. From what I've heard, AI is able to ape the style of a handful of famous composers but cannot yet at any rate recapture the spark which makes the better works from such composers immortal. Especially structurally, they seem to be all over the place.
I'm repeating myself, but this is further proof that the AI king has no clothes. The output is all over the place structurally because they do not understand structure. The algorithms is essentially just a probabilistic next-token predictor. I.e., given an initial segment of notes, it outputs the most likely subsequent note based on statistics computed from its training data. There is no high-level understanding happening here; any resemblance to high-level structure is merely coincidental, a shadow of the structure inherent in its training data. The algorithm is literally incapable of producing anything that isn't a part of its training data.
Effective musical structures arise from a deep understanding on the emotional impact of the notes as a whole, and the sequencing of this emotional impact to deliver an effective dramatic arc. Current "AI" has a loooong way to go before getting anywhere near the point it can even begin to approximate this.
Couldn't agree more with these points.
I hope you're right, but AI seems to be progressing like a California forest fire during El Niño. And just this morning someone on Reddit, who was joking, but still, said that I must have used AI to write my 572 book on timpani scoring. Like I said, he was joking and meant no harm, but who knows how many people are going to seriously think such things?
Composers of course know what AI can do as of today, but there are a lot of other people who hear about AI music who don't know, and, people being cyncial these days (as they have every reason to be), might begin to wonder how much of that tone poem is really yours. I can just imagine people at a pre-concert "meet the composer" q&a asking "How much does AI play a part in your music?" Argh!
And I think some people without the ability to write their own pieces that can stand on their own could probably be using it pretty soon to provide the initial foundations for a piece, and at most transcribe and rework and insert transitions and have something that is not solely their work, but mostly AI generated. Once someone does that and it results in a scandal of sorts, then suspicion will probably fall on the rest of us. But, I hope I'm wrong.
Wesley
just as well El Niño is ending, isn't it
But some good points there.
Perhaps in classical or contemporary pieces the AI is incapable of drawing a convincing or interesting structure. But . . . what about pop / mainstream music?
Structure in that regard is a piece of cake. I just "made" a convincing satirical pop-punk song in about 20 minutes that could easily pass for a radio hit from two decades ago. Intro, verse, chorus, verse, chorus, outro, done. With some DAW work I was able to effortlessly just copy and paste the "chorus" the engine spat out back in after the 2nd verse, now I have an incredibly commonly structured song that I promise you no one would listen to and say "that sounds fake" or "that's AI" (Except those here, because of course, now you know). Maybe I'll start an "AI band."
I argue that I do think the program understands structure to at least a simple capacity, because the transitions it came up with in my song work very well and seem to flow naturally. However, I want to clarify that I'm not suggesting the material sounds entirely too original, nor do I endorse using solely AI generated music for commercial content. My standpoint remains on the side that tries to avoid any naive views out of bias and inherent hatred of the tech.
You mean something like this? ;-)
I do not have any hatred of the tech. My main motivation is skepticism about the surrounding hype. The tech itself is, frankly, pretty impressive, given what it can do that just a few years ago no one would have believed (even though the underlying algorithms have been known for decades). However, it is not what the hype is making it out to be. There are very real limitations, inherent limitations, that are rarely acknowledged. And what really ticks me off is that naïve people are believing it, they're buying into the hype and treating AI as though it were some kind of miraculous human-level intelligence (or somewhere thereabouts), which it is obviously not. The hype sweeps these inherent limitations under the rug and the crowd is misled into believing it to be something it is not. That's what I strongly disagree with.
Dang, someone beat me to it!
But I was thinking more along the lines of starting a 'band' that churns out very generic but catchy and effective songs, all while there really aren't any living, breathing members in the band or even writing the music.
But hey, that's the reality for a handful of pop acts these days anyway, right?
See, you can one-up Compressorhead. They have robot players but human composers (and programmers); now you can start a competing band with both robot players and robot composers! ;-)