Checked out Udio after WAN show....WOW!

[KevanAU]

So I was watching WAN show the other day and saw the whole thing with Udio Ai song generation, as a guitarist and musician, I'm actually blown away with just how good this stuff is getting, I'm not sure exactly how I feel about it though, I had a bit of a play around and I got this to come out (see linked video), It's the heartbroken country song about a farmer who's Ai tractor done left him for another man, It's stupid, it's weird, and maybe given the direction we're going with technology, maybe it was a song that would have been written, but...also, maybe not, the songs about some guy that craps his pants at work, the Pilk song....

Do you think these things would ever actually exist without this kind of tech?, like I said, I'm a guitarist and musician, writing music for some of us isn't THAT overly difficult, it can be a pretty lengthy process though, I'm personally not a vocalist so while I probably could have written something like this, there's no way it'd have vocals like that without hiring someone to do it, then it's still a vocal track that needs to be added in and mixed.

There's a fairly decent amount that goes into writing and producing a song and I don't think people would always be willing to put in what it takes to produce an awesome version of something really pointless and dumb...but you still want to hear those pointless and dumb songs anyway though right? Is this a good use of this kind of technology?...I kinda think it is, I kinda love it lol.

 

[Biohazard777]

2 hours ago, KevanAU said:

... maybe it was a song that would have been written, but...also, maybe not, the songs about some guy that craps his pants at work, the Pilk song....

Do you think these things would ever actually exist without this kind of tech?

Just a few I remembered from the top of my head.
So yeah sure stuff like that would exist, far less of it of course, and perhaps not in every imaginable genre.

 

2 hours ago, KevanAU said:

There's a fairly decent amount that goes into writing and producing a song and I don't think people would always be willing to put in what it takes to produce an awesome version of something really pointless and dumb...

I dunno, hanging out with some punk rock bands when I was younger... lots of pointless and dumb songs hah, which didn't take much money or time to make.
Then again... punk... I'm sure it would have been a lot harder finding large ensembles to do the same hah.
Anyways, it sure has enabled a lot of people (with and without musical background) to make this stuff, quickly as well.

 

[Kisai]

It doesn't impress me. It creates maybe passable flavor-text/music you might hear in a shopping district inside a video game, but it doesn't sound real. It sounds like an AM Radio.

 

This is the problem with most "music" and "voice" AI, is that taken separately (See RVC) you can make a perfect clone of another song because all the AI does is "autotune" the voice B from the original voice A, but you use the same backing audio. All you've done is made a cover using the source voice, which to me isn't a "cover", it's akin to "nightcore"'ing a song where you just speed it up 50% and have done nothing else to it.

 

I'm not sure what the underlying process is for Udio because I've honestly just picked half a dozen different songs to check the genre adhere'dness but it seems like everything was washed with a noise filter that ranges from "radio" to "phonograph". I'd say most of these don't sound like what they claim to, and the ones that do, sound like they're inside a bathroom or hallway or down the street, or something.

 

They all lack "professional sound" mixing feel to it. If this was 1960, you could probably get away with it on LP or tape. 

 

Thinking about it for a minute, I believe I know what they did. They likely used commercial music as training data from different periods, because that would explain the incoherent levels of noise. The AI doesn't understand the "hiss" of a LP isn't part of the music.

 

[Needfuldoer]

22 minutes ago, Kisai said:

It doesn't impress me. It creates maybe passable flavor-text/music you might hear in a shopping district inside a video game, but it doesn't sound real. It sounds like an AM Radio.

That's generally how I feel about all this algorithmically-generated stuff. It's passable, disposable filler that falls apart under scrutiny, but you're not supposed to pay much attention to it in the first place.

 

I think the noisy output is trying to hide its shortcomings, like adding "film grain" to a generated "photo".

 

It still takes human input to get quality output from the plagiarism nexus, and even then its output works best if it's just one tool in the toolbox. (Just look at what "There I Ruined It" can do using voice cloning. The output isn't 100% generated; it's a human writing lyrics and singing, then running that work through an algorithm.) It still can't stand on its own. Generated images look "too" smooth, generated music sounds "too" phony.