Yes, were you able to resolve it?

Nope, still getting it. The controllable has been broken for a couple days now.

I’ve raised a support ticket. Think @hecko is on it.

am not on it

i don't have control over the talknet notebook

Can we tag the champ, @User ?

Seems like every notebook I’ve used, he ends up fixing it. I call’em The Champ. Maybe he will see this.

You're thinking of the HiFi-GAN problem

Just tried running the TalkNet Controllable and I get this:

It’s been doing that since March 8th.

noooooooo and I just finished my dataset too ; A ;

OH late to the party but have you tried `!pip install`ing dash? you might have already but if not it's definitely smoothed out a few module errors in training for me so it's worth a try!

speaking of errors tho this may be a silly question but for those of you who got past step six in the training notebook(s), how did you coax the cell into completing it? 👀 I tried juuuust about everything suggested in the stack exchange myself but I cannot for the life of me solve the logger error and other than a few shoutouts on the Lightning AI threads no one else seems to know what it is or how to fix it and none of the suggested alterations in there have made it budge >:/

There's a trail of dependency issues that I tried to follow. I don't think the solution is as simple as that

# !pip install dash
# !pip install jupyter_dash
# !pip install dash_bootstrap_components
# !pip install torch_stft
# !pip install kaldiio
# !pip install pydub
# !pip install frozendict
# !pip install unidecode
# !pip install pystoi
# !pip install g2p_en
...

When you mentioned that Colab had recently updated their python from 3.8 to 3.9, were you referencing this Talknet issue or the HiFiGAN one?

both

though with hifi-gan the fix was as simple as changing a number

You're definitely right about that, I've been running into that myself buuuut sometimes it's possible to get around the errors by playing along until the code seems functional, I got around step five that way. Course if there's OTHER errors popping up after everything's been installed... > u >;;

I haven’t tried that. Did that fix the controllable for you? The controllable is still broken, I just tried to run it.

I just don’t know enough to fix it. I’d have to ask my “coding friend”. I like TalkNet the best so I hope the controllable can be fixed.

I haven't even gotten to the controllable yet because I can't finish the training but it did allow me to get through the first five steps of that. What happens is every time you see a module error it's the notebook telling you that it needs something it doesn't have. What you have to do is go into the code and install the thing it's asking you to by hand and you can do that by using the

!pip install

command. ( hence the blurb at the end that talks about missing dependencies ) Now there have been other wonky things going on with the notebooks and a lot of it very probably has to do with inconsistencies in the software. What may happen is you get another error of a different kind because it's not running the version of the module that other parts of it need but you may also get lucky and find that all you needed to do was install a few missing components to bring everything together!

I feel that though, I was looking forward to using my first ever non-UTAU singing model ; 3 ; I hope the developers get everything ironed out soon so that we can continue using it!

Dang, sorry that’s happening to you! I’m not even using it for singing, I’m using it for talking, but it’s the only thing, that I know of, where you can use a reference track and it doesn’t blend the reference track in with the final one like Diff-SVC.

I trained my models years ago so I don’t know what’s happening with it now.

There used to be more ppl in here using it. Idk, if everyone switched to something else?

I imagine it will get fixed. They’re probably just busy, everyone here is a volunteer basically. So they probably just haven’t gotten to it yet.

I'm sorry you're in a jam yourself, it's never fun for us non coders to have to go in and figure out what the notebook is trying to tell us :') Talknet is a powerful tool that has a lot of really good stuff and you definitely deserve to not have to worry about the reference track 😬

YEAH I heard something about waiting for an update on Monday? I think what's going on is we're just holding our breath while the developers figure stuff out because I think some of what's going on are confirmed bugs

Oh, okay. Yeah, it’s probably something got changed or updated that the Colab is connecting to that’s messing it up. That seems to be the problem most of the time.

didnt justinjohn make a post in contrib