One focused utility

About NiceVois

NiceVois exists to make one awkward RVC workflow dependable: provide voice audio and leave with portable model files you control.

Why it was built

Training RVC models has traditionally meant maintaining a local NVIDIA environment, using third-party voice platforms, or cycling through notebook sessions that can break as dependencies and hosted runtimes change. This project moves that training pipeline behind a simple browser interface while keeping the actual model files downloadable.

What the product is

It is a cloud control plane for RVC v2 training. The web interface receives the job, external GPU providers perform the compute, and private storage transfers and retains the source and resulting artifacts according to the displayed policy. The public workflow can use Kaggle first and RunPod as a safe fallback for defined provider-availability failures.

What the product is not

It is not text-to-speech, an instant audio transcoder, a public voice library, or a tool for copying people without consent. Users must own the voice or have clear permission to create a reusable model.

Product principles

• Portable output: offer the .pth, .index, and complete ZIP instead of trapping the model in one interface.
• Truthful progress: show real stages and epochs received from the training worker.
• Recoverable jobs: continue after a browser closes and preserve validated models for the displayed window.
• User control: let users download or permanently delete inactive jobs in their private workspace.
• Permission first: require clear rights to the voice and prohibit deceptive or abusive use.

How claims are checked

Product claims on this site are tied to the implemented source and verified training runs. Technical references use the official RVC Project documentation where possible. Runtime and perceptual quality vary with the source recording and available cloud GPU, so observed test results are labeled rather than presented as universal guarantees.