Joshcore - AI generated covers of Null

  • Want to keep track of this thread?
    Accounts can bookmark posts, watch threads for updates, and jump back to where you stopped reading.
    Create account
I'm pretty happy with how this one turned out (considering the wacky source vocals and the minute amount of effort that went into it).
Ween feat. Jersh - Dancing in the Show Tonight:
 
Last edited:
This is a lot of fun.

Zoloft took a bit more effort with the mixing, and towards the end I just gave up and overlaid both vocal tracks (AI Jersh's faint voice cracks in the background of the harmonies makes it pretty brown imo, good stuff). I just imagine they're all in the studio and the rest of the band hopped in on the harmonizing. Not like anyone's ever gonna listen to the end of this track anyhow after that long organ solo, so it wouldn't be worth the tremendous effort of isolating and crossfading everything properly for that outro bit.

Ween feat. Jersh - Zoloft:
 
Last edited:
Big Mommy said:
Couldn't do much this week but everyone else knocked it out of the park. Could someone link me to the Fast Car cover from yesterday's stream? Also, don't think I posted these:

null_perfect_day.mp3
Lou Reed - Perfect Day
- Few times it hisses a little but I likes it!

null_complicated_postmoder_jukebox.mp3
Postmodern Jukebox (Avril Lavigne) - Complicated
- The last song worked well so I wanted another with clean vocals. This turned out a little hissy.

null_exit_music_for_a_stream.mp3
Radiohead - Exit Music (For a Film)
- Exit Music (For a Stream). It freaks out a lot. I had to rebalance the sound throughout. Intended to come back and fix it more but been busy.

null_please_forgive_me.mp3
David Gray - Please Forgive Me (Live)
- I like it. Clean vocals worked well.

null_going_to_a_town.mp3
Rufus Wainwright - Going to a Town
- Decent but goes a bit strange at points.

This was me trying to find songs with clean vocals that could be converted easily.
Click to expand...
Damn the Perfect Day cover is too good! That's a great outro tune right there.
 
Fulton J. Ween said:
This is a lot of fun.

Zoloft took a bit more effort with the mixing, and towards the end I just gave up and overlaid both vocal tracks (AI Jersh's faint voice cracks in the background of the harmonies makes it pretty brown imo, good stuff). I just imagine they're all in the studio and the rest of the band hopped in on the harmonizing. Not like anyone's ever gonna listen to the end of this track anyhow after that long organ solo, so it wouldn't be worth the tremendous effort of isolating and crossfading everything properly for that outro bit.

Ween feat. Jersh - Zoloft:
jershzoloft.ogg
Click to expand...

Superior version, I like how you covered the AI freakouts up with different effects. Nice!
 
Got a few new ones made

The Family Friendly Noose Song by Rusty Cage


Njet Molotoff
 
Ween feat. Ralph - It's Gonna Be a Long Night


Ween feat. Jersh - It's Gonna Be a Long Night


Ralph is great for shouty stuff. Jersh kinda works if you pitch him down (default pitch sounds like a troon), and distort to somewhat simulate shouting, but I think I overdid it.
 
Last edited:
Did anyone try to make Josh say the Chills copypasta? It's this:
Number 15: Burger king foot lettuce. The last thing you'd want in your Burger King burger is someone's foot fungus. But as it turns out, that might be what you get. A 4channer uploaded a photo anonymously to the site showcasing his feet in a plastic bin of lettuce. With the statement: "This is the lettuce you eat at Burger King." Admittedly, he had shoes on.

But that's even worse.

The post went live at 11:38 PM on July 16, and a mere 20 minutes later, the Burger King in question was alerted to the rogue employee. At least, I hope he's rogue. How did it happen? Well, the BK employee hadn't removed the Exif data from the uploaded photo, which suggested the culprit was somewhere in Mayfield Heights, Ohio. This was at 11:47. Three minutes later at 11:50, the Burger King branch address was posted with wishes of happy unemployment. 5 minutes later, the news station was contacted by another 4channer. And three minutes later, at 11:58, a link was posted: BK's "Tell us about us" online forum. The foot photo, otherwise known as exhibit A, was attached. Cleveland Scene Magazine contacted the BK in question the next day. When questioned, the breakfast shift manager said "Oh, I know who that is. He's getting fired." Mystery solved, by 4chan. Now we can all go back to eating our fast food in peace.
Click to expand...
 
I have very limited computer skills and even less free time. I humbly suggest that Planet of Weed by the Fountains of Wayne would be a fantastic Joshcore song, remade into "Planet of Sneed"
 
Finally got RVC webui and Demucs running locally through Conda, and this is heaven compared to wrestling with google colab rate limits and various free splitters.

Ween feat. Ralph - I Don't Wanna Leave You on the Farm


Johnny Ralph - Niggers Suck!
 
Spoke too soon.. Now I'm stuck with a CUDA out-of-memory error for some reason (the same thing happened back when I was messing with stable-diffusion), and every answer available online is to people who are trying to train new models. I'm just trying to do the inference part, and no matter what (changing algorithms, reducing filtering etc.), pytorch just produces this error:

torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 496.00 MiB (GPU 0; 5.80 GiB total capacity; 3 .73 GiB already allocated; 441.06 MiB free; 4.47 GiB reserved in total by PyTorch) If reserved memory is >> allo cated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

Anyone know what to do about this? Tried rebooting as that seemed to clear out the reserved VRAM when this happened in stable-diffusion, but even that didn't do anything. I'm on Fedora GNU/Linux with an RTX 3060 running the proprietary drivers (and everything worked just fine until it suddenly didn't). I had to set up a conda environment for RVC because I had to specify an older Python version (the one in the repos was too new), and I must admit I have no idea what I'm doing with that. Never really used conda before.

Edit: found the solution here: https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI/issues/649
I guess my GPU is just too weak :(
 
Last edited:
Ween feat. Jersh - What Deaner Was Talkin' About


My personal Ween favorite. Just a shame Null had a bit of a frog in his throat on this one, so Gener had to provide a lot of backup on the chorus.
Those layered vocals just mess up the ai too much man. Even after a lot of TLC (hopping between the original track and demucsed backing) it still only sounds meh, not good :(
 
You must log in or register to reply here.
Back
Top Bottom