I guess I'm fine with that. Although I'm using a separate notebook to use your code (and for my own code to build my bilingual voice cloning machine), I could make do with that approach

Does anyone know how many wav files I should make so it sounds great? Around 200?

I used around 100 for my Mario model, and it came out decent, not the best but you can understand what he says. I'm guessing the more the better

as many as you have the patience to get

@melmelmel225 Also it's not about as many samples after awhile but the quality of them be it mic quality and background noise to phoneme coverage.

Should I change the batch size if I have fewer than 30 clips?

probably, keep the batch size below or equal to the amount of audio clips you have

make it divide the clip count evenly if possible because if you have e.g. 30 clips and batch size 20 then the last 10 won't make a full batch and will get ignored i believe so in that case it's better to do batch size 15, or 30 if you can

recommend 15 if you can wait and want better pronunciation

So I started training the model from a checkpoint, but it just displays the checkpoint, but doesn't start training it

i mean from what i know bigger batch size is better almost always

but i may be wrong

that means your max epoch count is higher than the one the model has

I think I understand

if you're training on a smaller dataset than last time, note that the notebook only stores the iteration count and then divides that by batches per epoch to get the epoch count

Yeah

I just increased the ephocs to 2000 and it worked

so if you trained one epoch of 3022 batches and then made it a dataset of 2 batches it'll think 1511 epochs have passed instead

Thanks, works now

So I have a dataset for my language, but it contains different male and female speakers, can it still learn to speak my language?

it would probably get very confused try to extract the single speaker with the most speech or if your character's dataset is large enough and your language's writing system is straightforward enough you might just be able to yeet it at the english base model and have it work somewhat

I have a question, Anyone knows how i have to do the process of the voice? because i don't know if i have to send specific audios like specific word, or only put any audio from the artist

pretty much any audio, as long as it's not too repetitive and is split into roughly sentences (a few seconds long per clip)

oh thanks, so i can put vocals, words and syllabes right? and then also some phrases

preferably not syllables

the ai works best when trained on full phrases

aaaaa okok, i undestand it

thanks so much

I have one more question, when I am typing out my wav files can I use question marks or do I have to end all my sentences with periods no matter what?

nah you can use question marks and exclamation marks accordingly