The complete path, in order: prepare the audio, choose how long to train, understand the files you get back, and put them to use. Each guide stands alone, so start wherever you are stuck.
Before you train
What happens between upload and a ready-to-use model package.
DatasetHow much audio do you need?Duration targets, what to cut, and why preparation beats length.
Source audioAudio requirements for RVC trainingFormats, duration, isolation, and the problems that weaken a model.
Training strengthHow many epochs should you use?Use audio duration and urgency to choose a starting point.
Prepare and clean your audio
Two passes, three files: the lead, the harmonies and the instrumental.
Why it mattersSeparate the lead before an AI coverLeave the harmonies in and the model converts those too.
Training dataWhy denoising your dataset mattersA model learns the hiss just as faithfully as the voice.
CleanupWhy de-reverb matters for AI vocalsReverb is a room, and a voice model cannot tell them apart.
After you train
Upload a vocal and compare the normal result with the NiceVois cleaned version.
The differentiatorWhy AI vocals sound roboticConsonants have no pitch, so they are the first thing a model loses — and how they get put back.
Finished songChange the voice in a Suno songThe transparent stem, RVC, consonant-restoration, and remix workflow.
Model filesRVC .pth vs .index explainedWhat each output contains and why the package includes both.
Loading itHow to use your RVC modelWhere each file belongs, and how to run your first conversion.
CoversMake an AI cover with your modelSeparate, convert, set the pitch, and mix it back.
Model formatsPTH to ONNX explainedWhen conversion helps, what it drops, and when to skip it.
Coming from another service
Whether your old models are recoverable, and how to rebuild one.
Focused alternativeReplacing Weights.gg trainingWhat this utility replaces—and what it deliberately does not.
No local GPUAn Applio alternativeThe same RVC v2 output without an NVIDIA card or a Python setup.
Own the fileA Kits.ai alternativeWhen you need a portable model rather than a hosted voice library.
Custom voiceA Jammable alternativeTrain your own voice instead of picking one from a catalogue.
Real trainingA Musicfy alternativeA full RVC training and the model files, rather than a 30-second clone.
Case studyWhat Weights.gg got rightAnd why a free model catalogue could not last.
How we test
Measured training times and what comes out of them.
MethodologyHow we testThe setup behind every number we publish.
Upload permitted audio and download a portable .pth, .index, and complete ZIP.
Train an RVC model from audio