Actually, set the batch size to 44 and see what happens.

id rather use batch size 11 for that low amount of wavs

Will '-s make the AI malfunction?

You mean apostrophes?

yups for example I have lines like:" 'North , CathedralWest , Kemm MansionSouth , City Gates' " because it was ripped from the game, and they are said out loud by the narrator. But they have they are between apostrophes in the game files, and i don1t think ther is a regex expression that could remove them easily.

They would be okay, I think.

Okay thanks! I'll feed them to the AI and we'll see I guess.

If not I'll report the results here

what about -

probably okay

Retraining Prime 3.1 in Kaggle because Colab is just too slow and limited.

I could only get about 20 to 30 epochs in per session before getting disconnected.

I'd kinda like to get this update done before Christmas.

Oh yeah I have the same problem. Do you know the link to a trainer?

Remember when I said I was gonna update both Mermaid Man and Barnacle Boy? Well I'm updating Barnacle Boy first since the model of him that we currently have on the website sounds too monotone and strokes out on long sentences. Right now I'm separating each voice clip of him in Audacity.

Charge Yo Laptop

what's the best sounding voice ml?

Like what was eminem (pronunciation dict) trained on?

good question, one day i looked at the rapper section and boom it was there

I know 30Hz seems to be using that one

well it certainly seems that way

that would be a surprise, 30 makes his own models and i dont think he knows how to train arpabet models like that one

Oh well never mind then

Is 30 using reference audio?

dont think so, in the description of his videos he says he uses tacotron2, and tacotron2 doesnt have reference audio capabilities

Huh, thank you.

What's, in your opinion, the best sounding model all-around?

hmm alex trebek, juice wrld (rapping), the eminem model you just mentioned too

and those are tacotron2?