NNiceVois

RVC voice converter

Choose a voice or upload your RVC model, then add your vocal.

Voice conversion

Convert a vocal or full song with your voice

Loading allowance...
Three free conversions. Each counts once when it starts; listening and downloads are free.

Convert a vocal or a full song with a saved NiceVois voice or your own RVC model. Every conversion includes the normal RVC output and a NiceVois Clean Consonants version for more natural sibilance and breaths. Hear standard RVC versus NiceVois.

Voice modelChoose a saved voice or bring your own
.pth or model .zip
Your NiceVois voices
Or drop your RVC model here

Only upload a model you own or have permission to use. By uploading, you agree to the Acceptable Use Policy.

Pitch shift
Output Conversion results

Your converted audio will appear here

Choose a voice and audio on the left, then press Convert.

Summary

If you already have an RVC .pth file, you can use it here without installing anything. Upload the model, drop in an isolated vocal, and NiceVois runs the conversion on a cloud GPU. You get two files back: the normal RVC result, and a Clean Consonants version that repairs the robotic sibilance and missing breaths RVC is known for.

Two outputs from one conversion

Hear the consonants and sibilance change

Listen around the s, sh, f, and t sounds and the breath between phrases. Use headphones if you can.

“Guess that’s how it goes. Some things don’t work out. Used to chase your love, now I’m walking out.”

Normal output

RVC conversion

The direct result from the RVC model, unchanged.

NiceVois output

Clean Consonants conversion

The same converted voice with original consonants, sibilance, and breaths restored where RVC tends to sound robotic.

This vocal recording and the RVC voice model were provided and authorized by the NiceVois owner. Both files use the same source and conversion settings; only the NiceVois cleanup step differs.

New to this? Why AI vocals sound robotic explains what RVC actually loses, and why putting the original consonants back is the repair.

Bring your own model or pick a voice you trained on NiceVois. No GPU, no Python environment, and no local RVC install - the model file stays private to your account and is used only for your own conversions.

Use your own RVC model Hear normal versus cleaned RVC

Three steps

Convert once, compare normal and cleaned RVC vocals

  1. 1Choose your voice

    Select a NiceVois-trained voice, or privately upload your own RVC .pth or complete model ZIP.

  2. 2Drop in a vocal

    Upload the vocal, set the pitch shift if needed, and press Convert.

  3. 3Listen and download

    Compare the normal RVC result with the NiceVois cleaned version, then take the one you prefer.

Japanese AI singing voices you can use right away

No voice of your own yet? The converter includes the NiceVois voice library: AI Singer 御丹宮くるみ, AIみな〜ぎ, 夏目悠李, 小春音アミ and the official つくよみちゃん RVC model, each used with its creator's permission and credited on every result. Open the voice picker and choose the NiceVois voice library tab, or hear all five with their previews and credits.

Using a .pth file you already have

A .pth is the trained voice model itself - a PyTorch checkpoint holding the weights. It is not audio and cannot be played; it has to be loaded by something that runs RVC inference. That is what this page does, in the browser, so you do not need a local install to get value out of a model you already downloaded.

  • Upload the .pth on its own, or a complete model ZIP containing the .pth and its .index.
  • The matching .index is optional. It can help the converted voice track the original speaker more closely, so include it when you have it - see .pth versus .index for what each file does.
  • Models you upload are private to your account and used only for your conversions.
  • Convert only voices you own or have clear permission to use.

Upload the vocal, not the whole song

This converter expects an isolated vocal. Give it a finished mix and it will convert the instrumental along with the voice, which never sounds right. If what you want is to replace the singer in a complete song, use the whole-song route instead: it separates the song into instrumental, lead, and backing vocals, converts only the lead, and mixes the result back at the original balance.

What you need

You haveCan you convert here?
An RVC .pth fileYes - upload it and convert
A model ZIP with .pth and .indexYes - upload the ZIP
A voice trained on NiceVoisYes - already in your library
Only audio of a voice, no model yetNot yet - train a model first, then convert
A finished song rather than a vocalUse the whole-song route
Ready to use your own voice?

Bring an existing RVC model, or train a new one on NiceVois.

Use your own RVC model