Squidward Tentacles (RVC) 1000 Epochs

example

example 2 (real acapella used to convert)

who hurt bro lol

🔥

Full length version for the gamers.

Has anyone gotten the mangio fork working with crepe training? I can use any other method to train in the mangio fork, but crepe does not. It says the module doesn't exist, even with the folder in the directory.

This is the error preventing me from training.

check my buble post lol

The one about what errors you are getting?

It seems to be an issue with it failing to import torchcrepe, even if you git clone it into the RVC directory.

I already did pip show torchcrepe and it says its installed so this is extremely frustrating.

I got crepe to work, check dm's

We both got crepe working locally, I'm currently training a patchy the pirate model on crepe to 1k epoch. Hopefully it turns out good!

you got crepe training working locally?

i only have inference working

it gives me some syntax error

Here's my first local inference on my locally trained model for patchy, holy shit that sounds just like him! I tested it with an audio clip of patchy not included in my dataset. ( I didn't remove the music so it glitches at the end.)

I can show you how to get local training working, just hmu when you have about 4-5 minutes of free time, it shouldn't be too hard too fix now that I know how to fix it.

And here's one I did myself

interesting

is it just the torchcrepe issue? I had a different one where it seemed to complain about some syntax issue

Does it say which .py is creating the error?

I actually got that error at some point.

Not sure I asked gpt4 and got this as an answer: The error message you provided indicates a TypeError caused by a string formatting issue. The problematic line is: python Copy cmd = config.python_cmd + " extract_f0_print.py %s/logs/%s %s %s" % ( In this line, you're using the % operator for string formatting, but it seems that you're not providing the correct number of arguments to be substituted into the placeholders (%s). To fix the error, you need to ensure that you provide the correct number of arguments. Here's what you can do: Identify the number of placeholders (%s) in your string. In this case, there are 4. Make sure you provide the same number of arguments after the % operator. It seems the arguments are missing, so you need to add them. For example, if you want to replace the placeholders with variables a, b, c, and d, you can modify the line like this: python Copy cmd = config.python_cmd + " extract_f0_print.py %s/logs/%s %s %s" % (a, b, c, d) Make sure to replace a, b, c, and d with the actual variables or values you want to use in the formatted string.

I might just send you my RVC folder, as I have a lot of space and bandwidth on my google drive.

so what did you do to fix it?

But before that try replacing your extract_f0_print in your main dir and tell me what happens?

That was an error I think that happened to me along side crepe not being loading due to a module error. I know how to fix that error, but not sure what I did to fix that one.

I only edited these two pys to fix it.

this was a gradio error you fixed right?

or something else

Well, this was what I added to the beginning to get mine to work. # implement pip as a subprocess: subprocess.check_call([sys.executable, '-m', 'pip', 'install', 'torchcrepe']) Basically once it starts feature extracting, this code runs and makes sure torchcrepe and dependencies in the environment are working, once I clicked feature extraction, It installs that stuff first, however it glitches out after that and you have to close the python console and restart it after removing that line of code as it is only needed one time.

well for now, I'm gonna upload my Mang rvc folder for you to test and see if it works.

ill test this in a bit

if it doesnt work

ill dl ur folder

It's up if you need it:

Thanks for this, I'll try it out

Let me know how it goes!

idk which way is better, but chatgpt recommended to add 1 line to the infer web script which was a sys append to the rvc folder, and that honestly fixed everything since it helped identify where dependencies were installed

you posted a wall of code that I couldnt find in infer-web.py unless im dense

my bad, I just realized it goes to a different script. I'll be right back

that line of code goes right above in the "extract_f0_print.py" script

thanks

did they fix the memory leak stuff in the new commits today?

I haven't checked today's update, waiting on others to confirm if it's worth updating

I don't think it's fixed per the info on the webpage I read

laptop 3060, 6gb vram, 64 crepe size is training just fine on 5 minute dataset. and quickly too

1 epoch per 3s ish

weird as hell because on this same dataset with those same settings it was 5s per epoch, on an *a4000*.

something changed postively for sure

(and yes the models are exporting and inferring fine)

Squidward Tentacles (RVC v1) 1000 Epochs, Classic

patchy singing how bad can i possibly be sounds like a reality i will make when this is posted