There's so much data for Princess Peach...I just hope that there's a way to get HIFI-Gan in the system so the quality could get boosted...

with "in the system" you mean in uberduck? because uberduck is indeed using HiFi-GAN

Just be aware that Peach has had multiple VAs

Hewwo

what are u guys using here

wdym

to generate those images

I am using Tacotron2AutoTrim to trim a 12 minutes video. Can anyone tell me what values should I put in each line? "Enter Min silence Len...." "Enter silence thresh" "Do you want to skip large duration files"

Or leave them in default values?

Don't use this

The dataset will suck

Thanks for saving my time and energy

Aborting it now

I used this and it started speaking gibberish

And it wastes your colab gpu

if i do make a video using tacotron2autotrim ill show how to make it trim properly you cant expect it to know when sentences start and end and expect it to be accurate on words every time, it aint magic

honestly mega should update that colab to transcribe individual files too without cutting them, that would be amazing

Mine keeps speaking gibberish every time I train a model with autotrimmed dataset

What's this error? RuntimeError: CUDA out of memory. Tried to allocate 20.00 MiB (GPU 0; 14.76 GiB total capacity; 13.63 GiB already allocated; 13.75 MiB free; 13.71 GiB reserved in total by PyTorch)

Out of memory

Reduce either the batch size or the length of the longest clips.

Reduced batch size from 30 to 20. It's working now.

yeah just like i said it aint magic, it cant figure out when to put commas or periods

but atleast it did 70% of the job when making shadow so yeah that was handy

i want a alternative of the auto trim

okay, manual trim

Manual trim is tedious

I am so lazy lol

You'll ultimately have to do it. You need to check if autotrim did everything right.