cherrymint/rvc_webui repository overview

⁠Intro

As in the short description said - This repo is mostly meant to work on runpod.io⁠ <--- contains referal-code. I have tried it on my own PC aswell (NVIDIA GeForce GTX 1070 with CUDA 12.2) and it works. Your milage may vary tho...

⁠Tutorial

There is a newbie-tutorial available here⁠

⁠Tech- and Background-Infos

This repo uses an alternative Dockerfile. (Details can be found here: https://github.com/Husky110/rvc_web_ui-docker⁠)
There are two tags which I will update irregulary and one tag that is here only for archive-purposes.
A little bit of background: It seems like the original dev (RVC_Boss) has some weird standpoints on open-source-software in general (see https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI/issues/2109⁠ ), so I will keep his UI in it's own tag, but also provide the new repo made by fumiama.
All tags represent the state of their respected repo by the time the image was built (including all required models and downloads required to run) - except for the mangio-tag... I don't know from when this was - somewhere in early 2024 tho... :)
These are the tags provided:

• applio -> This is actually another project where one of the authors asked me to support it. Check them out here: https://github.com/IAHispano/Applio⁠
• rvc_boss -> The tag rvc_boss is for the original Repo at https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI⁠
• fumiama -> This is for the new repo at https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI⁠.
• mangio -> This is the repo: https://github.com/Mangio621/Mangio-RVC-Fork⁠ The mangio-fork seems to be abandoned and seems to have some issues in its' latest state.(see https://github.com/Mangio621/Mangio-RVC-Fork/issues/213⁠) However - I still have a working copy on my PC and the models it produces are still pretty decend. If you want to give it a spin - have a go here. This tag will NOT be updated tho! (Edit: This tag does NOT provide a Tensorboard! I will add that later on if I have the time...)
• test -> Do not use this tag! It is meant for test-purposes. :) It might work, but I use it to do stuff on the other tags. :)

This container comes with everything pre-downloaded, so it's ready to run.
If you use runpod.io:

• Just use one of my RVC-WebUI-Templates. They are the ones with cherrymint/rvc_webui in them.
• For your interfering and training I've set up some folders for you. You can use them, but make sure you change the web-uis default values accordingly:
  • /app/dataset -> Put your files for training a new model here
  • /app/audios -> Put all files you want to interfer here
  • /app/audio-outputs -> Meant for putting all outputs
• Port 7875 is for the WebUI, Port 7865 is for TensorBoard, Port 7895 is for the filebrowser
• For training, I would recommend using 3x RTX A5000 or 3x RTX A4000 with 24 CPU-Processes and a Batchsize of 12. That gives you 1 epoch at about every 7 seconds (with around 40 minutes of training-material), so 1000 epochs take about 2 hours, which will cost you 2-3$. Using newer or more cards did not do anything beneficial in my tests. (Written at 2024-06-10)
• To access the files, just use the filebrowser. You can also set it up with ssh - check the GitHub-Repo for that.