PG Music Home Forums General Discussion Help! Tech S.O.S (Off topic) Voice Changer

Sure.

Perhaps the most popular tool to use is RVC (Retrieval-Based Voice Conversion). The source code is available on GitHub, and there are a many tutorials on how to use it, as well as tools built to make it easy to install and run.

RVC can be installed locally on your computer, but it seems to have the most support on Windows, as opposed to Mac or Linux OSes. Once you've installed the RVC software package, it'll run locally as a web page. You can then do two things:

• Create a voice model by training RVC on audio; and
• Convert a source audio file using a voice model

It requires a lot of processing power to train a voice, so you're either going to need a computer with a powerful video card (to use the GPUs in the card), have a whole lot of time, run the training on a virtual machine, such as Google Collab, or hire someone to create the file for you.

On the other hand, converting an audio file using RVC can be done with a plain-old CPU. It'll even happen in real time if your computer's processor is fast enough.

There are a lot of tutorials out there, and even a number of forks of the various flavors of RVC. I haven't actually used RVC myself, but I'll supply links to what look like fairly reasonable tutorials. Cavaeat emptor, cave canem, etc.. You're a lawyer, so you love it when people write stuff in Latin, right?

I picked these particular tutorials because they appear to be targeting the most recent versions of RVC.

The simplest approach would be to use Google Collab. There's nothing to install on your computer, you don't need any GPUs, you can test it out free, and choose to pay when you need more processing done. What you will need is a Google account.

There is a web interface for RVC, and Google appears to be preventing Collabs with web interfaces from running in their "Free" tier. So there are a number of older RVC Collabs that no longer work in the "Free" tier.

Here's a video tutorial where someone's created a Google Collab for you. The video shows you how to run the Collab. It automatically installs RVC into the Collab. Once you've done that, you can load up training audio, train a voice model, save it to your Google Drive (so it doesn't go away when you close the Collab), and then apply the trained model to a source audio file. You can also retrain a model, and load a prior model:


Link to Google Collab: https://colab.research.google.com/drive/1hmKPNeeReO4NHzOktJFMI2plYKy5F7ZI?usp=sharing

Notice that he runs only 100 epochs of training to the model, as he's using a free Collab. It takes a lot of processing (and time) to get a high-quality voice. That's where paying Google for the Collab comes in handy.

I also noticed that he didn't split the training data up into a bunch of 10 second files, as most tutorials would have you do. It apparently speeds up the training time to use smaller chunks, but gives less coherent output. (This particular Collab's code seems to only accept one training file anyway).

This one Google Collab is all you need to create a high-quality RVC voice model, and apply that model to target audio. Easy peasy lemon squeezy.

If you want to run things locally, you'll need to install RVC on your machine. And you're likely to have the easiest time using a Windows PC instead of a Mac or Linux machine.

Here's a tutorial (among many) on how to do that. Again, if you don't have a supported GPU card, you might not be able to do any training (because some versions of the packages don't allow that), but you'll be able to perform conversions using a voice model locally:


Here's a tutorial on how to install a web-based RVC Voice Changer Client. It's aimed more for people who what to use RVC as a real-time voice changer, but it supports file input and output:


There's also free program called Replay that's uses RVC, but has a simpler UI. Here's a tutorial on how to install and use it:


Did that help?