Train an RVC voice model online
Upload clean voice audio and receive a validated .pth, feature
.index, and complete model package. No software installation or notebook setup.
- WAV, MP3, FLAC, M4A, and OGG
- Real stage and epoch progress
- Portable RVC files you control
Live cloud work
Active trainings
Up to two voice models can train at once. Additional uploads wait safely for the next slot.
Selected job
Training details
Epoch progress comes directly from the running RVC trainer.
- 1Audio receivedSecure cloud upload
- 2Dataset preparedPrivate training dataset
- 3GPU assignedWaiting for cloud capacity
- 4Voice model trainedNative RVC training
- 5Model packagedPTH, index, and validation checks
- 6Model readySaved to your library
An inference-ready early model is available while training continues.
It has been validated, backed up, and added to your model library.
The provider returned an error. Your other saved models were not affected.
New model
Start from an audio file
Persistent cloud storage
Your model library
Your .pth, .index, and ZIP remain available after closing the browser or restarting your computer. Source audio and temporary training files are removed after validation.
Checking private cloud storage…
Direct, portable output
From audio to an RVC model in three steps
This is model training, not a file-format rename. The service prepares your audio, runs RVC training on a cloud GPU, validates the artifacts, and returns files you can keep.
- 1Upload clean voice audio
Choose a supported file, name the finished model, and use the duration-based epoch recommendation.
- 2Follow real training progress
See dataset preparation, GPU queueing, live epochs, validation, and any available checkpoint.
- 3Download the complete model
Take the named .pth, the retrieval .index, or a ZIP containing the complete portable package.
Explore NiceVois
Tools and guidance for the complete RVC workflow
Go straight to the input you have, or learn how to prepare, train, and use the files you receive.
What happens between upload and a ready-to-use model package.
Dataset qualityAudio requirements for RVC trainingFormats, duration, isolation, and the problems that weaken a model.
Training strengthHow many RVC epochs should you use?Use audio duration and urgency to choose a practical starting point.
Model filesRVC .pth vs .index explainedWhat each output contains and why the complete package includes both.
Direct answerAudio to PTH explainedWhy the operation is model training rather than a file-format rename.
MP3 inputMP3 to PTH for RVCWhen compressed audio is useful and what quality problems to check.
WAV inputWAV to PTH for RVCHow lossless source audio moves through the cloud training pipeline.
No local GPUTrain an RVC model onlineWhat stays in the browser and what the cloud training worker handles.
Focused alternativeReplacing Weights.gg trainingWhat this utility replaces—and what it deliberately does not.
EvidenceTesting methodologyVerified artifacts, pipeline checks, observed runtimes, and limitations.
Previous NiceVois serviceLooking for lyric-changing services?The original personalized-song service is still available through Fiverr.
Common questions
Before you train
Is this literally converting audio into a .pth file?
It is training an RVC voice model from the audio. A .pth is the resulting model-weight file, so the operation requires dataset preparation and GPU training rather than a simple format conversion.
Which audio formats can I upload?
The current interface accepts WAV, MP3, FLAC, M4A, and OGG. Clean, dry voice audio with minimal background sound produces the most reliable result.
How much audio should I use?
The public beta currently accepts up to five minutes. Use as much clean, single-speaker material as the limit allows; recordings shorter than one minute may not produce a reliable model.
Do I need an NVIDIA GPU or local RVC installation?
No. Training runs on an external cloud GPU. Once training has started, you can close the page or shut down your computer and return later.
What can I download?
You can download the named .pth by itself, the .index by itself, or a ZIP package containing both. When available, an early inference-ready checkpoint can also be downloaded while the final run continues.
How long are beta files stored?
Public-beta model files are available for seven days. Successful source audio and temporary training files are removed after validation. Download the package before its displayed expiry date.