:|

Yes

That’s right 🙂

Try now?

Christ why do my models keep coming out entirely illiterate

I've got plenty of sentences each in there own files and the list clearly dictates everything

but each time I set it to train what I get back is terrible

Did you make sure it's mono and 2k

@UserShould be fine now

Just to be clear, somebody is going to look at those links and parse through how to complete the process, right?

Uberduck Tacotron 2 Training is not working it keep telling me runtime stop

Who is that from?

Uberduck

So far, I have 59 voice clips for Mr. Herriman. I'll keep getting more clips of his voice from more "Foster's Home" episodes.

Currently Working on my First Voice.

Far from perfect, but with more references. It'll be better.

I got the references btw. So no worries with help.

also if this isn't the right place to put tests in there, then I'll delete it.

huh

well it's a good first attempt

i think next time i'll let it train for longer and maybe expand the dataset juuuust a smidge if i can find more isolated liam vocals

So my main problem with the process is that I can't synthesise the dataset because I'm missing a file metadata.xml but I don't know where I'm meant to get this

Is there any other ways to make sure a voice you are training sounds good besides more audio files? I have a voice Im trying to work on that has 111 audio files and that is literally everything I can give it. But, the voice doesnt sound great and it has seizures with small sentences. Ive tried lots of ways to fix it, like splitting it by one sentence, and not using apostrophes. I have seen loads of models that use little audio clips and still are able to speak non gibberish like Principal of the Thing from Baldis Basics, only has around 20 voicelines but he still doesnt have seizures with small sentences.

i heard lowering the batch size to a minimum helps?

shouldn't it be maximum?

try both ig vsbdnfvmgbh

normally my batch size is at 24 with a K80 or 32 with a T4