I’ll still send the dataset just in case

ok

oof, I've done that before

Sent it as a drive link since the file size was too big

some of the files are stereo

e.g. 101.wav

Yeah I was worried that was the case

Probably the new ones I added which is 86 and up

Time to fix those files

There. My updated Yumi voice model is gonna be approved soon, because its previous version sounded awful. My Cheese voice model from yesterday didn't make it, so I've submitted it so it will be here soon.

How do the alignment graphs and the validation loss look? If they're still improving you could train longer. If not, it probably won't get any better

Will take a look at the new yumi and cheese later today

previous version had very coherent alignments (but still crispy), this new one is still 50/50 on good alignments and trailing off

I might have to cut up the dialogue more

validation is at .07 even if the alignments are bad

you guy don't add jenny wakeman to uberduck.

At long last I finally fixed all the clips and as of right now the model is being trained for the new files

Worth it

I can't really tell if maybe my dataset is unclean but I still get gravel in my generations

i should upload the dataset maybe and get it checked

@User I have submitted another voice model of a superhero who is... TOASTERRIFIC! If you could guess what it is...

I actually think the alignment is pretty good, I think your issue is just that that the vocoder is probably the LJSpeech vocoder so it doesn't handle deep voices. well. If you want to submit it we can try on my vocoder which was trained on deep voices, I suspect it will sound pretty good.

lol, is it brave little toaster?

i think it may be Powdered Toast Man from Ren and Stimpy

CORRECTION! Too bad Gary Owens, the voice of the beloved PTM, is dead.

how can i fix this?

Nope. It's Powdered Toast Man. Thx to @Echoes for guessing.

You have the transcript file as List.txt.txt and not clipper_train_filelist.txt

oh thanks, im an idiot