They made stickers?

Any luck?

0/0 [00:00<?, ?it/s] 0% 0/60 [00:00<?, ?it/s] --------------------------------------------------------------------------- AttributeError Traceback (most recent call last) in () 91 for r, _, f in os.walk("/content/wavs"): 92 for name in tqdm(f): ---> 93 convert_to_22k(os.path.join(r, name)) 94 95 # Convert to JSON in convert_to_22k(inpath) 41 if inpath.strip()[-4:].lower() != ".wav": 42 return ---> 43 ffmpeg.input(inpath).output( 44 inpath + "_22k.wav", 45 ar="22050", AttributeError: module 'ffmpeg' has no attribute 'input'

Dammit, man.

Guess, TalkNet isn't my thing.

Replace the whole "ffmpeg.inptu(inpath).output" with ffmpeg -i inpath -r 22 inpath +"_22k.wav"?

What else is after line 45?

Can you send screenshot instead of paste?

Or if youre using wav files at 22k anyway just remove that whole if statement on line 41 onwards

Okay.

Try just removing the if statement first

Can I do this later, I apologize but I deleted the tab because I lost hope and wanted to redo it later.

Yep np! I don't really know what Im doing but looked up ffmpeg and tried to work out what that code is trying to do. Other folk will know lots more anyway though.

What hardware would you guys recommend for training vocal synthesizer models locally? Trying to decide if I should buy google collab pro or just get a better gpu. I'd be looking into training multispeaker tacotran2 models and messing with TalkNet potentially

Seems like the code is broke anyway so best waiting for more official folk to have a look. Could even just be that a new version of a module has been released and breaks the code.

Totally depends on your budget. Collab is pretty handy as you don't need to install python and any other sort of packages, and the notebooks are easy to find and implement. But if you're happy coding and downloading extra stuff look up good Nvidia graphics cards in your range, its the most important. Then a good PSU obviously, get a load of ram, cpu and motherboard matter a bit less, as long as the MB has the slots for all the other components.

I've got no issues with installing or messing with code as that's a big part of why I'm doing this. Made a discord bot I'm adding TTS functionality to that supports other transformer models.

Yea tbh, I would get a Graphics card if I had the money and just cause Im interested in all sorts of other ML and resource intensive programs.

Mostly wondering worth of training time on the V100s/P100s vs. Something like a 2080ti or 3090

Probs worth getting one just so you're not limited.

If I get 2080tis I can pair up a few as well

GPU prices right now just suck

So debating if collab pro is worth it until they normalize

I mean colab is as fast as you're gonna get if I'm not mistaken. Pro usually lets me run on Nvidia Tesla V100s. But the perk of doing local is you can keep your pc on overnight without fear of disconnection. Colab pro does let you run stuff for way longer though.

Yea 24hr limit

Tho max of 16GB Vram

So you cant train large transformers

Ah fairplay

Do you know how intensive models like tacotran2+Waveglow and TalkNet+HiFiGan are on VRAM?