Is so-vits-svc a hosted service?

No, it is an offline toolkit that runs entirely on your own machine. You must provide your own compute resources, typically a CUDA-capable GPU, and there is no hosted inference service.

What are the system requirements for running so-vits-svc?

You need a CUDA-capable GPU, correct PyTorch and torchaudio versions, and must handle manual preprocessing steps like f0 extraction and slicing. Training can take hours to days per voice.

Is the so-vits-svc project still actively maintained?

The upstream repository was archived in November 2023, making it feature-frozen. Bug fixes and new features are now provided by community forks rather than the original maintainers.

Does so-vits-svc include a user interface?

The repository itself does not include a one-click UI or built-in interface. Users typically script the process or pair it with third-party WebUI wrappers to interact with the model.

What ethical guidelines does the project recommend?

The maintainers frame the project as intended for fictional or consented voices only. They explicitly disclaim liability for misuse, such as imitating a real singer without their consent.

so-vits-svc

SoftVC VITS Singing Voice Conversion — open-source pipeline for training and running singing-voice models.

Free· Free / open-source (AGPL-3.0). You provide your own compute (typically a CUDA-capable GPU) and training datasets.AudioSoftVC content encoder + VITS backbone + NSF-HiFiGAN vocoder; optional ContentVec, HuBERT-Soft, Whisper-PPG, WavLM encoders and shallow-diffusion module.

Visit website →

In short

so-vits-svc transforms existing vocal recordings to match a different singer's timbre while preserving pitch and phrasing. It is best for users who want a locally trained, fully controllable model and are comfortable with Python and CUDA setup.

Best for

Producers, researchers and VTuber/virtual-character teams who want a locally trained, fully controllable singing-voice model and are comfortable with Python, CUDA and dataset prep.

Skip if

Anyone who wants a hosted, click-to-clone service, needs a supported product with SLAs, or intends to imitate a real singer without their consent.

so-vits-svc (SoftVC VITS Singing Voice Conversion) is an open-source, offline toolkit for training and running singing-voice conversion models. Rather than synthesising speech from text, it transforms an existing vocal recording so it takes on a different singer's timbre while preserving the source pitch, phrasing and intonation. Under the hood it feeds SoftVC-encoded content features directly into a VITS backbone and swaps the standard vocoder for NSF-HiFiGAN, which cleans up the characteristic 'electronic' artifacts of earlier SVC systems. The 4.1-stable branch adds shallow-diffusion post-processing, multiple pluggable content encoders (ContentVec, HuBERT-Soft, Whisper-PPG, WavLM), static and dynamic speaker mixing, a feature-retrieval / clustering module for stronger timbre control, and ONNX export for downstream deployment. The project is aimed at researchers, vocal-synthesis hobbyists, indie producers and VTuber / virtual-character creators who want full control over a locally-trained voice model — dataset curation, preprocessing (f0 extraction, resampling, chunking), training loop, checkpoint management and inference all live on your own machine. There is no hosted service and no one-click UI in the repo itself; the community typically pairs it with third-party WebUI wrappers. Note: the upstream repo was archived on 2023-11-11, so it is feature-frozen — bug fixes and new encoders now come from forks. The maintainers explicitly frame the project as intended for fictional / consented voices only and disclaim liability for misuse.

Editor's take

Still the reference implementation for open-source singing voice conversion — the NSF-HiFiGAN + shallow-diffusion combo genuinely sounds better than most hosted 'AI cover' toys. But it is a research codebase, now archived: budget for a weekend of setup, a decent GPU, and be honest with yourself about whose voice you are training on.

— The AI Tool Bible editorial team

Pros

✅ Fully open source (AGPL-3.0) and runs entirely offline — no per-use fees, no data leaving your machine.
✅ State-of-the-art singing quality for its generation: NSF-HiFiGAN vocoder + shallow diffusion noticeably reduce breath and sibilance artifacts.
✅ Pluggable content encoders (ContentVec, HuBERT-Soft, Whisper-PPG, WavLM) let you trade off timbre leakage vs. pronunciation fidelity.
✅ Speaker mixing (static and dynamic) and clustering-based timbre control give producers real creative knobs beyond one-shot conversion.
✅ ONNX export enables inference on non-PyTorch runtimes and lighter deployment targets.
✅ Huge community: 28k+ GitHub stars, dozens of active forks, tutorials, and ready-made WebUI front-ends.

Cons

⚠️ Upstream repo is archived (Nov 2023) — no official updates, security fixes or new-model support; you rely on forks.
⚠️ Steep setup: CUDA GPU, correct PyTorch/torchaudio versions, manual f0 extraction and slicing, and hours-to-days of training per voice.
⚠️ No built-in UI or hosted inference — you either script it or bolt on a third-party WebUI.
⚠️ Documentation is uneven and partly Chinese-first; several config knobs (diffusion depth, cluster ratio, encoder choice) require trial and error.
⚠️ Serious ethical / legal exposure: cloning a real person's singing voice without consent runs into copyright, publicity-rights and (increasingly) deepfake-specific laws — the license does not absolve you.

Use cases

Singing voice conversion (AI covers)VTuber and virtual-character singing voicesCustom vocal timbre for indie music productionSpeaker mixing and timbre morphing experimentsVoice model training on curated datasetsResearch on VITS-based voice synthesisONNX export for lightweight SVC inference

Frequently asked

Is so-vits-svc a hosted service?: No, it is an offline toolkit that runs entirely on your own machine. You must provide your own compute resources, typically a CUDA-capable GPU, and there is no hosted inference service.
What are the system requirements for running so-vits-svc?: You need a CUDA-capable GPU, correct PyTorch and torchaudio versions, and must handle manual preprocessing steps like f0 extraction and slicing. Training can take hours to days per voice.
Is the so-vits-svc project still actively maintained?: The upstream repository was archived in November 2023, making it feature-frozen. Bug fixes and new features are now provided by community forks rather than the original maintainers.
Does so-vits-svc include a user interface?: The repository itself does not include a one-click UI or built-in interface. Users typically script the process or pair it with third-party WebUI wrappers to interact with the model.
What ethical guidelines does the project recommend?: The maintainers frame the project as intended for fictional or consented voices only. They explicitly disclaim liability for misuse, such as imitating a real singer without their consent.