if your dataset looks like

dataset.zip
| 1.wav
| 2.wav
| list.txt

Dream Island

Doesn't even looks like this

"Mario wearing a diaper" - Yeah, right...

interesting, i always thought it was me including list.txt sometimes w the wavs

It ended up being a goomba wearing a mario-colored diaper

Does anyone know the minimum seconds a wav file can have? Will one word sentences make it not work?

like 2 maybe

it must be a coherent sentence

try this too

Try different AI models from the main page

Queen Elizabeth II playing video games

uRAgYiKf2ptLf9AWGpUY (gone wrong)

i put in eddsworld and these are the results i got

you can see edd in the bottom left

"A drawing of a rubber duck with Uberduck written in the background"

35 minutes of data, and trained until validation loss 0.13, and sounds like this... bruh why i spent 5 hours waiting for this and now google hates me...

Has anyone used Colab Pro is it worth it ?

for uberduck only, its not really worth it unless you cant be bothered to wait longer for training, but colab pro is alright for only £8... colab pro+ on the other hand is a big scam and should not be bought at all

yes

"A drawing of a fusion of Homer Simpson and Goku"

All depends on what you need. We've def needed pro+ a few times for the high ram environments but I agree that the longer runtimes advertised is a total scam. Have had pro and pro+ colabs running side by side (personal/business) and the pro colab outlasted the pro+

Cursed

tacotron and talknet can easily run on K80's, which is what uberduck uses, so technically you dont really need anything. But Google hates free users now and its near impossible to use a lot of things without being forced to buy colab pro... but whats worse about colab pro+ is that some of their "good" gpus like the A100 doesnt even work well with certain notebooks from what ive heard from people who have bought pro+

Yeah I've def seen that as well. We had a multiclass classifier running the other day that had some kind of deep bug with GPU under pro+ wasn't an issue with pro.

Does anyone know how to fix the training steps 2500000 issue with Talknet? It says to delete the HiFi gan folder but it won’t let me because it says it’s not empty and only empty directories can be deleted

delete it from drive directly