It looks like the zip isn't uploaded correctly in this as it find only 1 file

It has the wav folder with some wav files and a transcript.txt in the dataset folder

Is it possible to just upload the dataset folder in the UPLOAD_HERE folder?

Yes, I do that when training models.

Have you not done that way?

I think I did one time, but it didn’t work out. I might try again in case I was tripping at first.

If that soesnt work then there's probably some really weird problem between you and colab I have yet to see

I know when I got that error ( albeit it was on talknet ) it was because it didn't like the way I wrote my transcript

Think that was the time I had to take the

wavs/

prefix out Though that wouldn't make sense here because I know legacy at least explicitly requires it to work

Any folder path works in pipeline iirc cause I think I tried that

wavs/ fishNuts/ I think they would all work

And I've looked through the lines in the transcript carefully and it seems just fine.

I MEAN TO BE FAIR it was in 2021 when those errors popped up for me and a lot has changed both with the notebooks themselves and the coding they run so it's very likely that my experience is outdated but that's really the only thing I can imagine it's taking issue with. I know it's a client-side bug but beyond that I'm not 100% sure

So I tried putting a dataset folder in the upload folder, and it didn't go in and gave me a red circle.

Wait...I just realize now. I think I need my wav folder to be a zip file! Lemme try that.

I could be wrong, but idk.

The

UPLOAD_HERE

does need a zipped dataset

Oh

Does the wav folder need to be zipped too, or should it stay uncompressed?

All items in the zipped dataset should be uncompressed itself

Dataset.zip
| wavs/
|   1.wav
|   2.wav
|   [...]
| transcript.txt

Right, okay. That's pretty much what I did earlier, but still couldn't get through this.

And then using the 2nd cell after uploading the zip gave me the list index out of range error where I'm always stuck at...

I also tried doing this the URL path way from a dataset folder, and it gave me some kind of a speaker 0000 error?

Okay, I think I got it now! It turns out I had to remove the * after putting the slash at the end of the dataset path! How dumb of me to think that lol

Well, now I just started training. This is intentional, right?

Ok nvm. I got the tensorboard to load

So I have the batch size be at 2 with the learning rate at 0.0002 cause my dataset has 19 wavs that are mostly over 5 seconds. Where do you recommend I should stop the training at?

This is my first time training this in pipeline btw.

Alright, I looked through the chat history so now I'll just stop until it looks good enough.

Man, I can't wait to see how it goes after it's done! Hopefully it's good... but I wouldn't be surprised if it is, cuz it is my first time training a pipeline TT2 model.