Snuffy

There's another snuffy model already, but that model didn't really sound exactly like her, so I made one myself.

Snuffy (RVCv2, 300 Epochs with 30 mins of data-talking model)

This is a model I primarily made as a talking model, and it can even laugh without sounding like a demon from another dimension!

30 mins of data, taken from a few random "highlight" videos. Not the main channel ones though, since they have a bunch of loud sound effects edited in.

And you can get a rough idea of how much data someone used by how large the index file is.

Based on that, this model has over 3x the data the other snuffy model has.

How would you even go about doing something like that? Changing a model for your voice in particular?

But yeah, I'll get em for ya.

In a hugging face link

https://huggingface.co/WhatTheHeck/HeckinModels/resolve/main/D_2333333.pth

https://huggingface.co/WhatTheHeck/HeckinModels/resolve/main/G_2333333.pth

I would be interested in trying out the end result of your extra training when you're done with it, if ya' don't mind.

Nice! A new snuffy model been wanting an updated one.

https://huggingface.co/WhatTheHeck/HeckinModels/resolve/main/Snuffy.zip

Voice data

Interesting, I didn't know you could put keyboard noises into it.

👍

Think I'm going to retrain this over the weekend with alot more "normal talking" data added on top, and some more laughing data. The highlight videos I used had her raising her voice a lot so I wanna channel more of her normal speaking voice.

Already like the voice a lot, but it could use some of that.

Hey, the updated model is now available to download in place of the old one. I won't say it's a lot better than the old one, but it's a little lower pitched, and channels her normal speaking voice better.

I recommend using a transpose of 10 if you're doing male to female with her, or -2 if female to female I guess.

Been messing around with merging models. Here's my snuffy model without any merging as a base.

And then it's a 50/50 split between snuffy and the jaiden animation model

100snuffy/50jaiden

100jaiden/50snuffy

Could you possibly make an RMVPE version? It'd be awesome.

Is it not RMVPE? I really don't remember.

Or how you even tell what method someone used to train their data lol

I mean as in make a separate version that's trained using the same Dataset here that was used but be trained with rmvpe. Like this. Mangio-Crepe (The current right now) RMVPE (The new optional link)

I know, I'm just saying I'm pretty sure I already trained this with rmvpe, unless I forgot.

Either way, I will retrain this when rvc v3 comes around.

Which is either in a month, or somewhere at the end of the yeard.

I've just seen a clip where snuffy got a vague donation message, wherein someone warned her that a streamer was using an ai model of her voice to stream. We got anyone who wants to step up? <a:KEKW:1022550661566574702>

~~just assuming they're using my model here, cuz that's funny to me~~