No local GPU or notebook
Train an RVC voice model online
You can train an RVC v2 voice model from a browser without owning an NVIDIA GPU or maintaining a Colab notebook. Upload permitted voice audio, follow the real cloud stages and epochs, then download the validated .pth, matching .index, or complete ZIP.
The browser acts as the control panel. Dataset preparation, feature extraction, training, validation, and packaging run on external cloud infrastructure.
What you do in the browser
- Choose a WAV, MP3, FLAC, M4A, or OGG recording.
- Name the finished model and review the duration-based epoch recommendation.
- Confirm that you own the voice or have clear permission to train it.
- Press Start Training.
What runs in the cloud
The service creates a private dataset, slices and resamples the audio, extracts RMVPE pitch and HuBERT v2 features, fine-tunes the RVC v2 generator and discriminator, exports an inference-ready checkpoint, builds a FAISS retrieval index, and validates the finished package. The public coordinator tries Kaggle first and can switch the same job to RunPod when Kaggle has a transient availability or quota failure.
Can you close the page?
Yes, after the upload and start request complete. The job is server-owned rather than tied to an open browser tab. Reopening the same browser workspace reloads its active and completed jobs. The interface reports the provider, current stage, completed epochs, elapsed time, checkpoints, and final download actions when available.
Current public workspace
| Capability | Current behavior |
|---|---|
| Audio duration | Up to five minutes per public-beta run |
| Epochs | Up to 100 in the public beta |
| Downloads | Individual .pth, individual .index, or complete ZIP |
| Retention | Seven days, with manual deletion available sooner |
| Local requirements | A modern browser; no local RVC or NVIDIA GPU |
Important: RVC is voice conversion, not text-to-speech. The trained model changes the vocal identity of a source performance; it does not independently generate words from text.
Use a voice you own or have permission to model.
Train an RVC model online