Audio & Music

So-VITS-SVC

Clone voices for singing

81 views

Visit github.com

Singing voice conversion beats regular text-to-speech in complexity. You can't just dump lyrics into a model and expect natural singing. So-VITS-SVC converts one person's singing voice into another's. Melody stays intact. Timing doesn't get mangled.

You get a visible f0 editor for pitch tweaks. Speaker mix timeline editor comes built-in. Real-time conversion works — if your hardware can keep up. Music producers working on covers or remixes can load a vocal track and morph it to match different artist characteristics without killing the musical elements that separate singing from speech.

So-VITS-SVC runs through web UI or Flask API. ONNX model export gets supported.

GitHub shows serious traction. 28k stars and 5.1k forks. Several projects built on top of it including MoeVoiceStudio and w-okada's voice-changer client. Models aren't compatible with standard VITS text-to-speech systems since singing voice conversion needs different training approaches.

Here's the problem though. Repository got archived in November 2023. No more updates. Version 4.1-Stable branch remains available but active development stopped. Developers who integrated So-VITS-SVC into their applications will need to maintain implementations without upstream support or switch to community forks that continue development.

Frequently asked

6 questions

Can I use So-VITS-SVC for real-time voice conversion during live performances?

Yeah, So-VITS-SVC can do real-time conversion -- but your hardware's gotta be up for it. You'll need a solid GPU or you're gonna get annoying latency during shows. Way more demanding than your typical voice effects.

What's the difference between So-VITS-SVC and regular VITS text-to-speech models?

They're totally different beasts. So-VITS-SVC models won't work with standard VITS at all -- singing voice conversion needs its own training approach. Regular VITS just does text-to-speech, while So-VITS-SVC keeps all the musical stuff intact (pitch curves, timing, all that).

Is So-VITS-SVC still being developed since the repository shows it's archived?

Nope, the main repo got archived in November 2023. No more official updates coming. The stable 4.1 version still works fine though -- you'll just need community forks for new features. MoeVoiceStudio and w-okada's voice-changer are keeping things alive.

How do I fine-tune the pitch and speaker mixing in converted vocals?

There's a visual f0 editor built right in. You can tweak pitch curves manually with it. Plus there's this speaker mix timeline editor that lets you blend different voice models throughout your track.

Can I integrate So-VITS-SVC into my own music production software?

Absolutely! You've got options -- web UI or Flask API depending on what you need. It'll export ONNX models too, which makes hooking it up to other apps way easier.

What kind of hardware do I need to run So-VITS-SVC effectively?

Get yourself a decent GPU, especially for real-time stuff. Exact specs depend on your model size and whether you're doing batch processing or live conversion. CPU-only? Forget about it for real-time -- it'll be painfully slow.