Voice of a female | 32k

So i made an original voice for the memes. It runs decently well with the voice changer (i havent done much testing as of writing) Download link: https://huggingface.co/Razer112/Original/resolve/main/Original.zip?download=true

with eq and reverb

without effects

Damn this is inpressive making an original voice from scratch. How did you make it? Just curious.

I just made several high quality datasets then trained them all together

IDK, is guma struhadlo na ceruzky (whom I have blocked) part of the UTTP or something?

Wait what?

If not, explain why you react like this, GSNC

Okay Idk why this person is like emojing my comments for stating an opinion on an voice model but okay.

Seems like they can't speak in this post cause the owner of this model I think has them blocked.

OK

Or maybe not. Either way I said what I said.

Here is a quick little voice changer test

So, what you did is mixing from 2 to 3 female voices in a same dataset?

Because this model sounds pretty impressive.

You could even sell it if you were a model master.

Yup

Can i talk to you on DMs?

If we are going to be talking about model making and such i would prefer that information be public

Oh, that's great.

I was going to DM you so this model post doesnt get flooded.

But anyways, which are the names of the voices you mixed and how much mins of dataset you used for each voice on the same dataset?

1. Remuchii with around 20 minutes of data 2. TheAlibee with 17ish minutes of data 3. Mookei with around 17 minutes of data

The voices should sound at least a bit similar to get good results?

Yea

How much voices are recommended to get a original one? 3 or 2?

3 or more

How much mins for each voice? 15, 17 or 20?

I did a test a bit ago with 2 voices each with around 20 minutes of data and the model turned out very robotic sounding

17 to 20

if one voice has more data then the others the model is gonna more sound like it

A good combination could be 17-17-20?

Mostly?

If you want the voice to slightly lean towards one of the voices, yea

Well, the goal is to make It sound original.

Then you would want all of the voices to have the same amount of data

That will only work properly when it's 3 voices and not only 2, right?

but do note that i havent made many og voices with 2 voices

So i could be wrong here

Can i mix any kind of voices like female Vtubers, videogame characters, ASMR female voices and streamers?

Example: Junko Enoshima + Calli + Kronii

I think you could test using 2 voices, one with 17 mins and the other one with 20 mins.

It's matter of testing and creativity.

That could work

Ill try that and publish the results

I mean, I've done multiple (from the same seiyuu) In fact, my Beyblade Burst Expanded Universe models are essentially that

I've been experimenting mixing voices in RVC as well and I'm thinking of making a video guide of it

The voice merging meta is becoming more known

The cheat code to make a good combined model is to use at least 3 voices that sound similar or at least kinda similar.

For example: Gura + Nanashi Mumei + Koseki Bijou (Biboo) can work.

You can make good merges of people with different kinda voices so long as they're the same gender

But like, I'm talking about merging models with like okada or rvc after you've already trained them, idk about combining them in the training phase.

but it seems kinda similar

I was talking about combining them in a same dataset.

Not W-Okada.

Ye, but I'm just saying the end result of this model here sounds similar in quality to the results you get from merging individual models

after the fact