shit i think i got that in the past, its because the face on the image wasnt detected i think

IT DID IT WITH IRON MAN AND THE MULTI PEOPLE

THE DEFAULT

come on

Well, I hope it works on CPU.

Man, I don't feel like downloading 250 MB

Someone fix the SimSwap colab when you can

SimpSwap

I am a simp for SimSwap

When you guys say your model's getting stonks, that's good, right?

Yes

Looks like I'm gettin' those stonks, Prime!

That's still early in training.

Anybody wanna play chess on it? Lol.

^ I believe it looks line that because I added Japanese Hiragana.

yeah unfortunately the model only accepts the latin alphabet and some punctuation

you'll have to either train on romaji, or use a japanese base model and ask zwf to add support for it

or if the rest of the voice is english then transcribe it like "toe doe keh teh"

Ah, wasn't aware of that.

Well I doubt it matters much when all chao speak is gibberish (・・)🥤

I have a question... is there a way I could test my model with a reference audio? After training it and testing it, I can only test it with a basic sound. Is there a way I could do it with a reference audio like in the main Uberduck page?

did you make a talknet model

Oh, I need to make a talknet model then.

exactly

Ah okay, thanks!

So for my machine learning let me think My problem is one that involves picking the right dataset with the right metadata files that will cause me no issues whatsoever when I try to build the machine How do I know which datasets in English and Japanese are right?

What's the best dataset collection for each language if I want to design a machine?

Well from what I've been seeing, most of what comes from it is audio with an associated

transcription

file that tells it what is being said In the assigned Audio. You might get more info from looking at papers and stuff, but it's probably easy to tell the language of the datase5 if you decide to open one of, given that the file structure isnt complicated. Or if you want to make your own that's also an option. Also depends if your making it to support only English, or for Japanese Hiragana/katakana.

As for the best dataset, idk, but there's probably a dataset out there that's more or less good. Or may have a lot of speakers in its collection