AI-created death metal

[Zer01]

https://motherboard.vice.com/en_us/...-channel-streams-ai-generated-black-metal-247

Based on the music of Archspire

 

[Musiscience]

This is genius. Mind = blown.

 

[narad]

Meh. Sad this gets any press as it takes away from legitimately interesting and novel things going on in AI music. Their research is taking some else's code and running it on Archspire :-/

 

[noise in my mind]

This is just terrible.

 

[GraemeH]

“Most nets we trained made shitty music"

First generation of this technology and they've nailed the genre 100%! Nice work.

 

[MFB]

Based on the music of Archspire

No thanks.

 

[Albake21]

Sounds fucking terrifying, and not in the good way.

 

[Descendant]

Honestly this sounds like Deicide played on 2x speed.

 

[KailM]

At least this is better than Djent.

 

[jonajon91]

This project is actually WAY more interesting than it's getting creddit for in this thread. They've done this experiment with a few different albums from Dillinger, to battles, meshuggah or the beatles. I think there are nine or ten albums/eps now and they lay each song out as the AI learns so the early tracks are more noise/experimental but the album becomes more clear as you listen to it. Therefor anyone jumping into a few seconds of the first track and dismissing it are not actually hearing the AI work.

The most successful ones from the album album "Diotima" by Krallice since black metal is all atmosphere and no substance (I kid), the meshuggah one, and the psyopus one for just being terrifying.

I think there are about a dozen albums cut and most of them are on this bandcamp.

https://dadabots.bandcamp.com/

The beatles one best shows the progression of the AI and tbh I wish the other albums followed the same formula.

 

[noise in my mind]

Yeah, the beatles one is terrible

 

[c7spheres]

Honestly this sounds like Deicide played on 2x speed.

I just played it on half-speed. Nope. I'll stick to humans. The only good AI thing that will exist will be robot sex dolls and slave workers. They don't talk back, You don't have to take them on dates and spend money on them, they don't eat, they don't age, and there always DTF!. I know, I need help. Or do I?

 

[M3CHK1LLA]

Yeah, the beatles one is terrible

yeah, i agree.

the only thing i could make out was "i am the walrus" or some such gibberish...

 

[777timesgod]

Disappointed, I was expecting a death metal parody from Weird Al Yankovic. I thought the title meant AL (Al) and not AI...

I am on the fence on AI music, I do believe that AI will give us human the boot in the future so music is a secondary consideration. Then again AI can help us with improving our lives so its promotion is inevitable.

 

[noise in my mind]

Disappointed, I was expecting a death metal parody from Weird Al Yankovic. I thought the title meant AL (Al) and not AI...
View attachment 68781
I am on the fence on AI music, I do believe that AI will give us human the boot in the future so music is a secondary consideration. Then again AI can help us with improving our lives so its promotion is inevitable.

AI creeps me out. I think we are opening pandora's box. I think we are fucked.

 

[narad]

This project is actually WAY more interesting than it's getting creddit for in this thread. They've done this experiment with a few different albums from Dillinger, to battles, meshuggah or the beatles. I think there are nine or ten albums/eps now and they lay each song out as the AI learns so the early tracks are more noise/experimental but the album becomes more clear as you listen to it. Therefor anyone jumping into a few seconds of the first track and dismissing it are not actually hearing the AI work.

The most successful ones from the album album "Diotima" by Krallice since black metal is all atmosphere and no substance (I kid), the meshuggah one, and the psyopus one for just being terrifying.

I think there are about a dozen albums cut and most of them are on this bandcamp.

https://dadabots.bandcamp.com/

The beatles one best shows the progression of the AI and tbh I wish the other albums followed the same formula.

I don't think there's anything interesting going on here in the AI. It's a kind of interesting model, but it's still a generic one, and not one that these guys had anything to do with. Running it on famous bands isn't itself research, and random listeners hear that the model generates locally coherent bits of songs people know, and get excited. That's just local level model bits over-fitting. None of these are getting interesting, repeating, high-level song structures, or true learning of what makes this music interesting at a song or album level. Frankly, the paper never should have been accepted anywhere.

I look forward to the day someone actually looks at genre-specific failings of current models, and proposes interesting extensions to better capture black metal / death metal, etc., in their differences from the piano or pop melodies that are usually the testbed for this type of research.

 

[Demiurge]

It's very weird. Y'know how when you're at a bar where music is playing in the background but not loud enough, so it becomes a nondescript shuffle of rhythms, vocal sounds, etc.- it's like that but at full volume.

 

[noise in my mind]

I don't think there's anything interesting going on here in the AI. It's a kind of interesting model, but it's still a generic one, and not one that these guys had anything to do with. Running it on famous bands isn't itself research, and random listeners hear that the model generates locally coherent bits of songs people know, and get excited. That's just local level model bits over-fitting. None of these are getting interesting, repeating, high-level song structures, or true learning of what makes this music interesting at a song or album level. Frankly, the paper never should have been accepted anywhere.

I look forward to the day someone actually looks at genre-specific failings of current models, and proposes interesting extensions to better capture black metal / death metal, etc., in their differences from the piano or pop melodies that are usually the testbed for this type of research.

No thanks, I'd rather humans produce the art.

 

[narad]

No thanks, I'd rather humans produce the art.

Well hybrid approaches are also interesting and more feasible. The AI makes suggestions, humans iteratively select/curate. I mean, for the most part, music differentiates itself from a lot of other art in that the process is not appreciated or considered an integral part of the piece. Seems a natural foothold for AI-assisted creation.

 

[noise in my mind]

Well hybrid approaches are also interesting and more feasible. The AI makes suggestions, humans iteratively select/curate. I mean, for the most part, music differentiates itself from a lot of other art in that the process is not appreciated or considered an integral part of the piece. Seems a natural foothold for AI-assisted creation.

Ah, I see what you are saying. I mean you can kinda do that with a random note generator.