Latest on MJJC

A.I Michael - opinions

Beano Wild said:
Also, if anyone has any multis (or acapellas) not public that would work to train the ai, contact me on twitter or something, I believe it's against the rules to transfer files like that on here. I'm planning on expanding the ai's training set from 35mins of audio to hopefully 1-1.5 hours.
Click to expand...
Would filtered acapellas help? The ones on the Acapella/Instrumentals thread sound clean enough.
 
Beano Wild said:
Hey guys, on a separate note before talking about this -
glad to be a mod here, hoping I can help make this a better place for you. And for me. And the entire hu- okay sorry.

But back to the topic, I have spent the last week researching and building an A.I model of Michael jackson singing, and after almost 40 hours of training the ai is scaring me at how it has reached the level of a really good Michael impersonator, and I'm only about 30% done training.

Now before you slander me, I will justify my actions by saying that if it wasn't me, it would have been made by someone most likely named "MJ_LEAKS2008" and would have been used to create fake "leaks". I am keeping the model completely private, and will only be working on songs either Michael covered but are unreleased (Delilah, Hot fun, Jane is a groupee) or songs michael himself loved (Automatic, Upside-down, How deep is your love, My Perogative). Another thing I'd love to try is to finish Hollywood tonight by using the written bridge.

Now, the reason to why I've made this thread - should I releass songs online with this tool, or keep them all private? And also what are your opinions on this matter, and do you believe Michael would have been fascinated by this technology?

Again, I would love to read peoples opinions (mostly so i have something to read on my work break, lol), if you want to criticise me go ahead too btw as I'd love to hear everyone's opinion.

Thanks lads (and again, I hope I can help ye all in the best ways I can!)
Click to expand...
just out of curiosity, how does one become a moderator here? what are the qualifications? does it pay?
 
I support it. Like it or not ai is the future. Look how much its changed the art field.
Would you be willing to give a breakdown of the training process? I think if more people understand the tech it will be easier to spot any misuse. I think any leap forward should be open source. (You dont need to share the MJ model though) I'm just curious on the process.
 
Beano Wild said:
what do you all think of this? i think it sounds nearly 1:1 with michael which creeps me the fuck out.
https://twitter.com/i/web/status/1629275247345119233
Click to expand...
As the AI gets better, it actually seems easier to tell the difference. This doesn't feel like MJ still, because there is a bit of a flaw, in that it almost sounds "perfect". MJ's warmth. His humanness. That made a lot of his songs. Like the way his voice notably cracks on Earth Song. It that's quality that's missing with auto-tune/melodyne, and AI. So this is very good and quite impressive. But I'll still always miss MJ, and we are not just in a world where MJ's absence is just okay. It's simply, a really cool advancement in technology to look at with interest.
 
wow! alot of questions and thoughts!
Mister_Jay_Tee said:
Yo. I want it.

Gimme Michael Jackson at home, I want it
Click to expand...
let me just check what the brunei royal family paid, give me a second..

ChanceToDance said:
Now I wanna hear an AI cover of Eaten Alive lol
Click to expand...
i did one myself but it sounded terrible, will prob work with this or another impersonator for that song once i make a gritty michael model.
Robbsaber01 said:
I support it. Like it or not ai is the future. Look how much its changed the art field.
Would you be willing to give a breakdown of the training process? I think if more people understand the tech it will be easier to spot any misuse. I think any leap forward should be open source. (You dont need to share the MJ model though) I'm just curious on the process.
Click to expand...
i dont know, i really dont want this to be used to make "new songs" and songs michael himself didnt have an involvement in. i do agree with the leap should be public, but i just dont know if i should let it known to the masses just yet as im still experimenting. will definetally make a video on it at some point though. and as for fake leaks, i will say that the highends on the ai will sound robotic as hell, so if new "leaks" comes out that attempts to mask the high ends and nobody can verify it, dont believe it.
wonderouzmj said:
No joy!
Click to expand...
i might do that one lmao, the soft model prob works best for songs like that and GOM.
Mister_Jay_Tee said:
As the AI gets better, it actually seems easier to tell the difference. This doesn't feel like MJ still, because there is a bit of a flaw, in that it almost sounds "perfect". MJ's warmth. His humanness. That made a lot of his songs. Like the way his voice notably cracks on Earth Song. It that's quality that's missing with auto-tune/melodyne, and AI. So this is very good and quite impressive. But I'll still always miss MJ, and we are not just in a world where MJ's absence is just okay. It's simply, a really cool advancement in technology to look at with interest.
Click to expand...
exactly. im doing this out of curiousity of what the technology can do, and it seems to be doing pretty damn well. i will say i cherrypicked that section of the song as it was prob the most Michael sounding part. other parts sound a little like malachi, but thats no fault of the impersonator, hes a great guy and hes done great work lmao.
 
Didn't think I'd ever hear rick astley covering Billie Jean yet here we are :ROFLMAO:
 
Beano Wild said:
i will say i cherrypicked that section of the song as it was prob the most Michael sounding part. other parts sound a little like malachi, but thats no fault of the impersonator, hes a great guy and hes done great work lmao.
Click to expand...
Hey, that's still way more than I ever expected. Could I maybe hear the whole thing anyway? I liked Jason Malachi's voice too lol.
 
Shouldn't this technically work for creating higher quality vocals out of low quality sources too? Like Slapstick for example.
 
Beano Wild said:
i might do that one lmao, the soft model prob works best for songs like that and GOM.
Click to expand...
Can you do this next...I really wanna hear yet I feel damn guilty for wanting to hear manipulation of Michael Jackson. It's like being comfortable listening to Jason malachi. I remember in 07 I used to tell ppl & get mad because they kept saying mamacita & etc was mj & im like....u really about to argue with me about this? U really think that's him?! (Sigh) but you know how that goes
 
It would be interesting if the first songs to be made with the AI are songs that were meant for MJ but he never recorded/weren't released. Since you can only do "soft" songs are you able to do Willing And Waiting, Joy and I Have This Dream?
 
ChanceToDance said:
It would be interesting if the first songs to be made with the AI are songs that were meant for MJ but he never recorded/weren't released. Since you can only do "soft" songs are you able to do Willing And Waiting, Joy and I Have This Dream?
Click to expand...
He said Willing And Waiting wasn't possible due to the harmonies and overdubs iirc
 
Beano Wild said:
every song is possible, id just need to rerecord them lmao
Click to expand...
How about doing the chorus of I'm Dreaming one day? will.i.am sang it pretty clearly on an interview so it's possibly doable.

Or hell, even the third verse of BoJ if Korgnex takes too long to leak it lmao
 
You must log in or register to reply here.
Back
Top