loop all files in directory

you don't have to do numbers even

for file in directory, append the name to a file

dir /B /O:N > list.txt

in command prompt

then open with notepad++, find and replace, turn on regular expressions (that or copy the contents into regexr) find

^.*\.wav

, replace with

wavs/\1.wav|\1

Won't that just name everyting 1.wav?

nono it doesn't rename anything

it keeps the filenames the same

it just takes the list.txt and turns

Something blah.wav

into

wavs/Something blah.wav|Something blah

I see. While I'm here, what does one give for a "val" dataset?

for talknet

usually the same file

if you wanna get fancy (and have a bigger dataset) you can give it just the first few dozen lines to speed up quality control, or even remove those lines from the training data to make sure you're not overfitting on the test data but you don't have enough audio to worry about that

What exactly is the val dataset? I was thinking it was like BasicSR where it's a test image, but now I'm confused.

it's used to calculate the loss value

Does anyone have a new link to the waveglow pretrained model from this google colab page? https://colab.research.google.com/drive/1AjsNJ42vnymL_gNXqG5xiEHt4lXAW_-C#scrollTo=Ps5c9xMChCV6 The current link is dead.

Is there a list of unicode that Talknet/Tacotron recognizes? For example if I want my transcripts to have "…" or "~" in them to denote trailing off or joking

use my synthesis colab

its already stored to a new pretrained link

Awesome, thanks so much!

looks like 1977 Chuck E. Cheese is developing more s t o n k s

Some random Master Chief dialogue from halo 3.

Was gonna do ||Super Ghostbusters Joel||, but after training and tweaking the dataset 2 times... yea he ain't happenin.

How long is the timeout period for collabs for GPU usage?

NVM, just buying colab plus lol

Also is it normal for the training graphs to look like this for the first several epochs?

My apologies for the dum dum questions, just new to all of this.