This RVC test measured two permitted voice recordings from job creation to a packaged, backed-up model. Submitted 31 seconds apart, the concurrent RunPod-first jobs finished in 4 minutes 34 seconds and 5 minutes 42 seconds.
Voice A finished in 4 minutes 34 seconds. Voice B finished in 5 minutes 42 seconds. Each job produced a portable RVC v2 .pth, matching .index, and complete ZIP package.
Run details
| Run | Source audio | Epochs | End-to-end time | Verified output |
|---|---|---|---|---|
| Voice A | 45.5 seconds | 50 | 4m34s | 54.9 MiB PTH, 5.5 MiB index, 60.5 MiB ZIP |
| Voice B | 56.84 seconds | 50 | 5m42s | 54.9 MiB PTH, 9.1 MiB index, 64.0 MiB ZIP |
The model names and source recordings are not published. Both completed packages passed the production artifact checks and were copied to verified private backup storage.
What end-to-end includes
The clock starts when NiceVois creates the job. It includes upload handling, dataset preparation, GPU assignment, RVC training, index creation, artifact validation, packaging, and verified backup. It stops when the completed package is ready in the user's library.
What this benchmark does not claim
This is a specific observation, not a universal speed or quality guarantee. Runtime can change with audio duration, epoch count, cloud capacity, and provider conditions. We verified the workflow and artifacts; we are not making a perceptual audio-quality claim from this test.
Why NiceVois is simpler
Traditional RVC training involves installation, dataset paths, preprocessing, feature extraction, GPU settings, and model packaging. With NiceVois, the customer uploads permitted audio, names the model, chooses the epochs, and presses Start Training once. NiceVois handles the training workflow and returns the portable files.
Your first voice training is free. Later jobs show a fixed workload quote before they start. Use only voices you own or have clear permission to train.
Train an RVC voice model