- Forum Clout
- 927
I made an RVC2 model of our favorite roadkill faced pedo. RVC takes an existing or real-time vocal audio and replaces the voice with the subject of the model, AKA voice cloning. The model isn't super dynamic, so it can't do impressions or whatever stupid voices he'd make. It's only trained on ~11min of 2013 gun rant audio, since it was the easiest way to get a clean block of him talking without interruption. It took about 30min to prepare the audio dataset and 1hr of training on a cloud machine, which cost less than a dollar. If this picks up any interest, I'll actually spend a little time on it and make a much better model.
Nana RVC Model
Finally addressing the rumors:
quick comparison:
My actual voice sounds nothing like him.
The downside is that I don't have a site or simple executable where you can just run this easily. There are a few ways to work with RVC, but I prefer this:
Retrieval-based-Voice-Conversion-WebUI
It takes some understanding of how to get code from github and run it in a python environment, but all the steps should be laid out in that readme. There are also tons of tutorials on youtube. It has a webUI that processes audio files, as well as an app that lets you process mic input input for real-time usage. If you use this, put the nana.pth file in the "weights" folder, then create a folder called "nana" in the logs folder, then drop the index file into that folder.
Also, keep in mind that this will only replace the tone and timbre of the voice, but not the style or performance. You actually have to talk/sing like the person from the model. That's why most of those AI vocal swap music videos on youtube sound so unnatural.
Officially, this is meant for research purposes only. It is not meant to be used for nefarious purposes...
Nana RVC Model
Finally addressing the rumors:
quick comparison:
My actual voice sounds nothing like him.
The downside is that I don't have a site or simple executable where you can just run this easily. There are a few ways to work with RVC, but I prefer this:
Retrieval-based-Voice-Conversion-WebUI
It takes some understanding of how to get code from github and run it in a python environment, but all the steps should be laid out in that readme. There are also tons of tutorials on youtube. It has a webUI that processes audio files, as well as an app that lets you process mic input input for real-time usage. If you use this, put the nana.pth file in the "weights" folder, then create a folder called "nana" in the logs folder, then drop the index file into that folder.
Also, keep in mind that this will only replace the tone and timbre of the voice, but not the style or performance. You actually have to talk/sing like the person from the model. That's why most of those AI vocal swap music videos on youtube sound so unnatural.
Officially, this is meant for research purposes only. It is not meant to be used for nefarious purposes...
Last edited: