VoiceStudio
82KDownload

Speech generation · Opt-in

CosyVoice 3

Multilingual zero-shot cloning with instructed speech, installed in its own Python environment.

  • FunAudioLLMBy
  • 10Languages
  • 24 kHzOutput
  • ~5.4 GBDownload
Get VoiceStudioEngine docsFunAudioLLM/CosyVoice

Can do

  • Voice cloningSpeaks in a voice copied from a short reference clip.
Languages10
10languages
  • Cantoneseyue
  • Chinesezh
  • Englishen
  • Frenchfr
  • Germande
  • Italianit
  • Japaneseja
  • Koreanko
  • Russianru
  • Spanishes

Runs on

  • PyTorch, isolated

    isolated
    PyTorchCUDACPU

    One-click install into its own Python 3.10 environment.

    Engine doc
Hardware
EngineCUDAROCmIntel ArcApple SiliconVulkanNPUCPU
PyTorch, isolatedSupportedNot supportedNot supportedNot supportedNot supportedNot supportedSupported

Supported Needs a build Experimental

Licence review

Preview registry disclosures. These source records do not establish permission for a complete model selection, its dependencies, or your use of a person's voice. A paid VoiceStudio licence covers the app's own code and does not replace upstream model or output terms.

Commercial model use: Not verified for this selectionCommercial output use: Not verified for this selectionModel redistribution: Not verified for this selection

Select a source below for its declared licence, revision, evidence and unresolved checks. An engine can offer several sources or variants; these records are not an installed-file manifest.

FunAudioLLM/Fun-CosyVoice3-0.5B-2512 · Evidence not verified
Declared licence
apache-2.0
Credit / notice
FunAudioLLM/Fun-CosyVoice3-0.5B-2512
Commercial model use
Not verified
Commercial output use
Not verified
Model redistribution
Not verified
Voice rights / consent
Not verified
Evidence revision
29e01c4e8d000f4bcd70751be16fa94bf3d85a18
Configured runtime revision
29e01c4e8d000f4bcd70751be16fa94bf3d85a18
Evidence checked
2026-10-03
Original evidence SHA-256
Not captured; original document bytes are not verified

Registry note: Licence identifier is upstream model-card metadata, not verified permission for this distribution. Component terms and attribution text require review.

Upstream publisher (new tab)Original evidence / model card (new tab)

  • Required components and their combined terms are not fully reviewed.
  • Exact downloaded files have not been verified.
  • Original terms have not been verified as a complete document bundle.
  • Provider access and any upstream agreement remain separate.

Registry 2026-10-09.draft1; evidence checked 2026-10-03. This is disclosure-only foundation work. It does not mean released apps enforce review before every download, and no real model workflow is verified ready by this snapshot. A local acknowledgement does not grant rights or complete a provider's own agreement or gated-access process.

Snapshot provenance and full inventory
Registry SHA-256
626f87eb82dfe114fe00f08ebc4754cb6522d23368d5206654a0ccdb6dd8f2d2
Export SHA-256
a47e3e0c421a39e1c16a0dfb79fd4224bbfe5d2ea8b254c02b7fa2e4b53fe7a3
Inventory audit base
06c6e077f0fc35149efefc3561e9be5ae835d916
Export source commit
090cd5b58cf771def46f64352a0de72c2432fc56

The licensing export has its own app source pin. The capability snapshot below uses a separate, older pin; neither identifies your installed app release.

Pinned app export (new tab)Download full registry disclosure JSON

The full export also preserves unresolved dynamic assets, including selected tokenizers, voices, alignment, watermarking, separation and language packages.

VoiceStudio is AGPL-3.0 and doesn't relicense model weights. Licence guide

Related

From the VoiceStudio app at 0834c8b. 27 models listed.

Quick answers