Looks like google collab has banned RVC voice models. I see several paid software options but have not tried them. I made several RVC models last summer but I dont see a way to create them now for free. Any ideas?

 

Hi,
I use RVC-WebUI (open source). It can be a bit touchy to install and use it as there is not much information but it works pretty well and generate good models.
https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI
Just tested the latest version, it's easier to use

 

Last edited: Jan 20, 2024

Thanks!

 

Is there a tutorial for this?

 

@daddytang
Sadly, no good tutorial.
But this is rather simple:

• Download the version for your GPU and extract
• run go-web.bat (your web browser will open the app page)
  
• Select "training" tab and enter a name for your model
  
• Specify your dataset full path (wav or mp3 files are OK, maybe other formats as well)
• most default settings are fine, set your epochs number
• Click the "one click training" button bottom right
• wait till the end

Once done you can add epochs and train more your model.

You can test your model in inference tab.

Of course training can take time depending on your dataset length, and CPU / GPU.

 

rvcDisconnected colab
terrible "code" quality but works

• colab limits gpu usage, 24h model training interval recommended
• dataset: normalized mono wav file named as file1.wav, processed with UVR-DeEcho-DeReverb, (available on mvsep, 10min max) stripped out of silences, optionally 16bit (sake of file size), packed to zip file and uploaded to Google Drive
• dataset length: less than 25 minutes on new pretrains else more than 10
• samplerate: 48kHz if old pretrains else 40kHz
• crepe hop: 64
• pitch extraction algorithm: rmvpe
• cpu threads: 2 if not colab pro else more (unconfirmed)
• batch size: more than 7
• pitch guidance recommended
• epochs: more than 40 on new pretrains else more than 100
• enable autosave, you can download models later (/content/Mangio-RVC-Fork/weights)

also reminder that replay is not a voice cloning app

 

Last edited: Jan 21, 2024

thank you for this!

 

Why is my Index file not being created? Constantly gives an error! Help!..
Sorry for Google translator.

 

Assuming one has a paid Google Colab account, are there any functional and comprehensible RVC model-training colabs anymore?
I cant get past the interfaces on either desktop apps or figure the colabs out, they all seem poorly ported from Chinese and largely incomprehensible. I succeeded in training a So-Vits-SVC model last year but there doesnt seem to be either a comprehensible tutorial or a working inference anywhere. Need help if anyone can get me over the geek gatekeeping walls!
Thanks

 

This is the best colad for online training:

https://docs.aihub.wtf/rvc/cloud/rvc-disconnected/

 

Great! And what about actually applying the model to target audio, what is the best interface or colab currently?
I dont have a local NVIDIA GPU unfortunately, but do have paid colab

 

https://docs.aihub.wtf/rvc/cloud/applio-colab/