The AI Music situation is wild and I feel like it popped up overnight

[FedUpWithTech]

For reference- I generated this for in about 2 minutes.

https://suno.com/song/34276095-affb-4dd8-80cc-b137284f44dd

What are all your thoughts?

Edited April 19, 2024 by FedUpWithTech
broken link paste :/

[BentleyOwen123]

that's erie, it lacks emotion but i bet it's only a matter of time until they find a way to add inflection on words and things, and an ai signing about tax write offs isn't something i thought i would ever hear

[FedUpWithTech]

1 minute ago, BentleyOwen123 said:

that's erie, it lacks emotion but i bet it's only a matter of time until they find a way to add inflection on words and things, and an ai signing about tax write offs isn't something i thought i would ever hear

Yeah I've spent a couple hours tonight just messing around with various AI music generators - its really crazy. My favorite thing so far is to take copypastas and make them into songs

[GuiltySpark_]

Funny tech to mess around with, but unless its a human person with talent and creativity singing or making the music, I simply do and and will not ever care. Waste of time.

[FedUpWithTech]

44 minutes ago, GuiltySpark_ said:

Funny tech to mess around with, but unless its a human person with talent and creativity singing or making the music, I simply do and and will not ever care. Waste of time.

That's totally fair- I'm afraid we are already too far gone as a society to combat AI. I'd imagine within 5 years most of these AI tools be it image generators, video generators, music, text, and whatever next becomes a big deal all becomes so good and indistinguishable from reality it will be impossible to navigate the digital world without AI help. I guess it really is as a good time as any to go outside and touch grass (and support local music)

[GuiltySpark_]

37 minutes ago, FedUpWithTech said:

That's totally fair- I'm afraid we are already too far gone as a society to combat AI. I'd imagine within 5 years most of these AI tools be it image generators, video generators, music, text, and whatever next becomes a big deal all becomes so good and indistinguishable from reality it will be impossible to navigate the digital world without AI help. I guess it really is as a good time as any to go outside and touch grass (and support local music)

Sure, and its going to keep growing, but normal people who follow artists and care about real music don't care.. and I don't see them ever caring. For now its just for tech kids and bros.

[Mark Kaine]

as a volaloid fan i can only shrug at this new "invention", sorry. 

[Kisai]

Most "AI Music" stuff is terrible, and it comes directly from the underlying tech not understanding what makes music "Music"

 

Music is not simply a cacophony of sounds smashed together.

 

Basically we've seen three different strategies:

1. TTS pushed through what is essentially vocaloid + midi

2. RVC, which autotunes one voice to the melody of another, without consideration of the underlying musical accompaniment. You will see these pop up as "AI covers"

3. Lyrics and music generated BY the LLM in the same way Dall-E/Stable Diffusion do "artwork"

 

None of these sound pleasant except number 2, and number 2 only sounds passable when the voice being autotuned uses a professional singer in the first place. But why would you bother with it? I've seen it largely used to change the singer of a song as an "AI cover" but it makes you ask "why?", it's like nightcore and LoFi, just speeding up or slowing the song down doesn't fundamentally change the song, but some people do prefer a certain listening experience. So maybe there are people who want every song sung by their favorite singer who has never sung it. It's still GIGO (Garbage In, Garbage Out.)

 

If you use someone's karaoke performance as the input for RVC, the output is still going to sound like that karaoke performance, mistakes and all.

 

Number 1 really only works for Choral or Rap. It will sound terrible outside of that, because fundamentally a TTS isn't generated from song data. But if all you need is a choral accompaniment, it'll do.

 

Number 3, which is what we've seen in the last few months, requires having been trained from the MUSIC and LYRICS together. So the result is that you're going to have a lot of lazy, mass-generated poor quality music because the people generating it, aren't artists.

 

The same people who mass produce "AI Artwork" en masse on deviantart. 

 

They're trying to say they're artists, when they're clearly not, and can't even tell why the things they produce lack artistic merit. While, yes, the AI's will improve to a point, they will plateau as a they consume all the input media that they are allowed to, and start feeding off each other and eventually introduce high levels of "bad" art. Which again is why labeling AI generated materials is a matter of survival for AI.

 

Anyway, Music is hard to begin with, to create stuff that sounds pleasing to the audience. AI will never understand this. Lyrics can be substituted (see Weird Al) to the same melody and still be a song, but an AI doesn't understand that.

 

 

 

 

 

[thevictor390]

9 hours ago, Mark Kaine said:

as a volaloid fan i can only shrug at this new "invention", sorry. 

It's not really analogous in my opinion. It's a digital replacement for your voice the same way you can synthesize a guitar or drum track. You still have to manually control every aspect of it, from pitch to timing.