Spongebob Squarepants (157 Epochs, RMVPE, 40K)

[feral badger]

157 epochs, i know it does not sound like much, but with almost 100 minutes of data its a lot (96 minutes to be exact) trained to exactly 40k steps RMVPE what else could you want? What cahracter should i do next?

This model was trained on my rtx 4080 super

Download Link: https://huggingface.co/ThatBlondeGuy/Spongebob-157-Epochs/resolve/main/spongebob157ep.zip?download=true

Enjoy 😉

[gusty fiber]

Rate the model

To rate this model, either use /rate or click the buttons below.

[feral badger]

Samples:

Graphs:

[dire bluff]

https://tenor.com/view/i-saw-w-gus-fring-gus-gustavo-deleted-gif-25440636

[feral badger]

posted them both at once 🤣

[dire bluff]

why does g and d total have strange lines around 25-35k and 36k(?)

is it a rvc bug? always trained models and never got these weird lines

[feral badger]

so i went back epochs, and it went back steps, i tried to get the model a better understanding of the voice

Went back to the latest checkpoint

Tried to nail it directly on 40K and was sucessful

[dire bluff]

you should try to train one time and then stop when g/total starts rising up

if im not wrong

to prevent "bugs" like those

but yea the model is still good tho, 100 mins of data is good to have a realistic ish model

[feral badger]

Im trying to figure out how to stop training without closing cmd, and saving the latest checkpoint

[dire bluff]

what are you using to train

like, what rvc fork?

[feral badger]

gpu or rvc application?

[dire bluff]

rvc application

[feral badger]

Mainline

[dire bluff]

try applio

it has stop training button, idk if other rvc fork has the stop training button

[feral badger]

which one do you use, or would recommend, just applio or somenthing else?

[dire bluff]

i use original rvc (for rmvpe) and applio (for mangio-crepe)

[oak rock]

is only a visual bug lol

[dire bluff]

i know its a bug lols

[feral badger]

i figured, because it did not taint the model

[feral badger]

does applio have rmvpe

[dire bluff]

ye

[oak rock]

technically is possible to fix it but i dont remember how xD

[feral badger]

lol

[dire bluff]

i was used to use original rvc due that it was a clean version of rvc with no edits but it didnt have crepe or mangio crepe lol

[oak rock]

applio training is faster than mainline, model results are the same

[feral badger]

really?

[oak rock]

yup

[feral badger]

is it easy to retrieve the files tho?

[oak rock]

locally? Yea theyre in the log folder of your model

[feral badger]

ohhhh ok just like mainline, and one more thing, does it have model inference and work with tensorboard?

[oak rock]

yep everything

[feral badger]

oh ok im switching, lol

[oak rock]

Remember to install the compiled version xD

[feral badger]

lol

I thought ai hispano got hacked?

[feral badger]

to be correct it is the newest version at the bottom?

[oak rock]

Yep

[oak rock]

Only blaize discord acc was hacked

[feral badger]

ohhhh

[oak rock]

And he recovered it

[feral badger]

thats good

[oak rock]

yep applio itself was fine

[feral badger]

when i saw the ping, i was like oh no, and left their discord server

[torn breach]

What sources did you use for the voice data?

It sounds like The Cosmic Shake

[feral badger]

i used some from the cosmic shake, like 30 minutes and the rest were other games

[torn breach]

What were the other games?

[feral badger]

i used tons of soundtracks i could pull from other old spongeboob games

[feral badger]

i dont remember, i know there was just like 5 or 6 sources, lol

[torn breach]

Ok

[feral badger]

@torn breach i used some soundtracks from the cosmic shake, the sb movie game, spongebob patty pursuit, bfbb rehydrated, thats all i could remember

there was like 2 or 3 more

[jovial hatch]

I still have not found a Spongebob model that can scream

can this one?

[feral badger]

idk, lets test it 🤣

do you have a screaming audio sample or somenthing else

[jovial hatch]

Diamonds is usually a good song to test that :3

[feral badger]

gotcha

[dire bluff]

SHINE BRIGHT LIKE A DIAMOND🗣️🗣️

[jovial hatch]

I hear plankton when reading this

[dire bluff]

i remember plankton everytime i hear this song

https://tenor.com/view/miau-adobe-after-effects-glass-breaking-default-preset-gif-5777596260912161139

[feral badger]

yea you could use my plankton model 🤣

[feral badger]

same

i just got it

here we go

[dire bluff]

@feral badger a question, do you use uvr locally?

[feral badger]

yes

[dire bluff]

cool

it feels good to have a rtx 4080

[feral badger]

yea it does

[dire bluff]

tho i'll buy a rx 580 in the future

to inference locally, use shaders in minecraft and use w-okada (if i get a virtual cable)

[feral badger]

the 4080 is super fast, i will say that

[jovial hatch]

hmm he doesn't scream sadly, did he have any screaming audio in his dataset?

[dire bluff]

fr

[feral badger]

yea a lot of it

[dire bluff]

rvc is limited to laughs and screams iirc

[feral badger]

let me try to clean the audio better

[dire bluff]

nah its clean enough

[feral badger]

ok lol

[jovial hatch]

https://tenor.com/view/i-saw-what-you-deleted-cat-gif-25407007

[dire bluff]

good for ya

[jovial hatch]

laughing definetly but I've had some models with good screaming/yelling

[feral badger]

i will probably make a screaming spongebob model soon 🤣

[jovial hatch]

lmao all screaming

[feral badger]

that would be hilarious

[jovial hatch]

that's odd, my scientist model I posted just now has a lot of screams too but doesn't do it often at all

maybe because most of the dataset is talking

[feral badger]

i dont think rvc is advanced enough yet to include true emotion, thats my thought.

[dire bluff]

x2

[jovial hatch]

agreed, my Oogie Boogie model has some super good range in his voice but still is hard to get some stuff out of him

[gaunt parcel]

Ayo? @jovial hatch level 35 !!!

[jovial hatch]

https://tenor.com/view/true-feel-that-thats-true-gif-7410050725709839510

[feral badger]

yea i noticed that

[jovial hatch]

I think mixing singing and talking helps models a lot

[feral badger]

it does, because in most cases the character will just be talking, and wont be screaming or singing.

[dire bluff]

🧽 bob 🟫 👖

[feral badger]

spongebobby 🧽

[jovial hatch]

true

[dire bluff]

spongebobby squarepanty 🧽

[feral badger]

Mr. 🟫 👖

[dire bluff]

mr. squarepants

[oak rock]

114 messages

[dire bluff]

yea we love chattin

[feral badger]

arrgh matey, my sponggyy

[feral badger]

🤣

i wonder if this is a record 🤣

[oak rock]

actually not, this causes rvc to merge both singing and talking in the final result and can potentially cause unstable gradient

[oak rock]

good luck xD

[feral badger]

💀

[jovial hatch]

oh

[feral badger]

they just need to give us a special role for posting spongebob models 🤣 all i do is post spongebob models

[analog urchin]

well damn

I have an rtx a4500 and it has 20 gb ram

but an rtx 4080 super ain't that bad

[feral badger]

awesome, thats crazy

the 4090 is so overkill