Thanks

Currently training the Zim model for uberduck.ai

the fnf ones are pretty bad- Ima try to make better ones- I got the music composing sofware KawaiiSprite uses just for this project

Uhh... just so ya know, we ain’t looking for beep boop skadoop. We actually had to dig through the sources to get actual lines. (Spooky Month for Skid and Pump, and FNF for Tankman and Monster).

If you’re that desperate, make a UTAU or something.

i mean, they only do beep and boops, you can't just make them say words

This.

about utau: i need to make how to do pronouns on japanese because if you make say AAAA and IIIII, they only say is AHHH and AYYYY

or smth like that because depends american and canadian is a strong accent

I said this before somewhere but UTAU is all about phonetics

I’ve seen someone make a utau out of a fucking piano. No. Really.

I’m not fucking kidding.

cv making be like, yeah... needs make models to say some kanji stuff

You gotta break each word into their ss-ah-oo-nn-d to get realistic pronunciation It is a thing for people to make Banks out of instruments but the quality is generally not as good as human models and certainly can't speak human languages

I wasn’t talking about human models when I made the UTAU remark.

AH

gotcha

🤔 damn i thought i've spent alot making cv with my cupcake voicebank with sentence mixing

to make it sound japanese

Depending on the size of the voice bank I can believe that! I had a multi-pitch bank made out of my own voice before I lost it and that took a good long while to build

wonder if that's possible to make vcv with ai models

Yyyyes but I'm not confident it'd sound... super great UTAU, like Tacotron, is very picky about its background sound and audio clips with the robotic quality to it will affect how well the program is able to read them, at that point you'd be better off stringing sentences together using clips from the dataset and feeding those directly into Utau. Additionally because it's a singing synthesizer program, you need to have your model sing with a straight tone for the best musical product and I am not aware of any speaking models that produce that kind of monotone

You can try it, I just wouldn't set my expectations super high tbh

well eh... i could do pitching monotone with melodyne/newtone

but i'm worried about pronounciations

like i said, american/canadian have a strong accents and they may say it something incorrectly as it is not japanese kanji stuff

That is where my previous comment comes into play

You'll need to input it phonetically and understand how the words are supposed to be pronounced

So like, "watashi" right? Phonetically it would read like "wah-tah-she" so both when naming files and when you're laying the notes down in the project, you'll need to name each note after the way it sounds in the corresponding language