Spongebob (80 epochs, RMVPE, 40K, RVC v2)

[latent lotus]

171 minute long dataset

trained on my rtx 4080 super

Download Link: https://huggingface.co/ThatBlondeGuy/Spongebob-New-80-Epochs/resolve/main/spongebob80ep.zip?download=true Enjoy 😉

Details:
Model Name: Spongebob
Model Creator: ThatBlondeGuy
Epochs: 80
Steps: 39280
Model Architecture: v2
Sampling Rate: 40000
Batch Size: 8
Pitch Guidance: True
Dataset Length: 02:51:17
Creation Date: 2024-10-26 04:23:35
Hash (ID): db7202f6a1c288f03d945a6aa5ea0af4608ca8bae63287646eda16d87920151a
Overtrain Info: Embedder Model: NoneMax Speakers ID: 1

[wicked bough]

Rate the model

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

[latent lotus]

samples coming soon

Graphs:

Sample:

[split igloo]

batch size? :p

[latent lotus]

8

[split igloo]

put it in the model description just in case

[latent lotus]

i just did

[split igloo]

ty

[gusty perch]

Woah, nearly 3 hours of voice data!
What sources did you get the voice data from?

[split igloo]

probably all the voice lines where the actor of spongebob voiced spongebob

[latent lotus]

all kinds of games, especially really old ones, to really capture his original voice

i went all out on this one, but i could not get past 80 epochs

[gusty perch]

Nice
My SpongeBob model is similar in that regard
It's a blend of his pre-movie voice (circa 2002/2003) and his post-movie voice (2004–2013)
I was able to train it on a 104-minute dataset for 500 epochs with an A100 GPU; although, for some reason, it sounds it could use more training or it should be trained on a better pretrain like Titan

[latent lotus]

are you talking about your model, like talking about adding a better pretrain

[gusty perch]

I'm thinking I should have trained my model using the TITAN pretrain rather than the original pretrain

[latent lotus]

ooohhhh gotcha

[split igloo]

titan has distorted harmonics y'know

[gusty perch]

What does that mean?

[latent lotus]

it works fine for me

[latent lotus]

same here bro

[split igloo]

that the model's spectogram will be kinda unstable iirc, distorted high-ends and low-end noise

*what i am trying to explain is dis

[latent lotus]

gotcha

[gusty perch]

You forgot to smooth the graph to 0.999