For: @Cyanicide Dataset by: @Cyanicide Model download: https://huggingface.co/Coolwowsocoolwow/Blaze/resolve/main/Blaze_e20_s660.zip?download=true

Transcript: My flame can penetrate anything! Let's go to the store. You have to use my power to stop it before it's fired, or use your power to grab it.

Blaze (Sonic 06) (GPT-SoVITS) (20 Epochs)

Bro @Cyanicide. You gotta stop beggin kimid for models everyday homie.

This is honestly so friggin good. Is there some tutorial I can follow to make a model like this? I've never tried GPT-SoVITS

Its pretty easy once you get the hang of it

If you look in https://discord.com/channels/1159260121998827560/1228880226432319548 you can find the colabs

Sweet So I'm seeing one for training & the other for inference. Which one is for what? & is there some instruction thing I could follow, like a video?

Training one is just for training and the inference is the one you have to use for inference

"Inference" as in?

You make a folder in the colab folder thing on the training one and upload your audio and copy the path and paste it into the thing

Using the model

I see; thanks for that

If you want to use this model you just copy the huggingface link and put it in the think where it says to paste the link into that

Assuming its already installed

And then you have to download the reference audio and put it into the files in colab and copy the path and transcript and put those into their respective boxes

So is there a way to use the model off of colab (maybe on a mac)? I know there's usually a 2-3 hr time limit on those links, for GPU usage

I think so but you would have to use docker

docker?

Yeah whatever that is

It just needs some commands I think

Or you could try running the .sh file in the GPT SoVITS folder

Where would I get the reference audio, like for this model?

Here

It's just 9 seconds? Don't models usually need like 10 min worth to sound alright?

Well its a TTS model thats been fine tuned so all it needs is that

Cool And would I have to change around any of the pre-set numbers on the colab links? (GPT settings, etc)

Not really it usually works with the normal settings for temperature and other ones

Ok, so does it matter which folder the reference audio gets uploaded onto? Or just stick it on a folder & copy-paste the path?

You can just upload it into the files but you dont need to put the reference in any folder just copy the path of the audio file and paste it

Understood Also, I'd prefer using flac files, but does training/reference audio/whatever other stuff require mp3 format?

It can be MP3 or WAV or FLAC