Skip to main content

Clone a voice

Train a custom voice or import a model

Instant clone uses only about 20 seconds of vocals.

Your voices

Tap a finished voice to use it.

AI Voice Cloning

Clone any voice with AI

Turn a few clean vocal clips into a reusable AI voice. Instant cloning gives you a zero-shot voice in seconds, while Best quality trains a dedicated model for the closest possible match — all in the browser, with nothing to install.

Already have a model? Import it straight from a Hugging Face or Google Drive link. Every voice you create lands in your library and the cover studio, ready to sing, rap, or narrate.

At a glance

Voice cloning at a glance

Setup
Runs in your browser
Speed
Instant or trained
Sources
Upload or import
Output
Studio-quality covers
Overview

What is AI voice cloning?

Voice cloning builds a model of a voice from sample audio, so it can re-sing or re-speak anything you give it. On VoiceDub you can clone instantly from a clip, train a dedicated model for higher fidelity, or import a model you already have.

  • Clone from as little as one clean clip
  • Instant zero-shot voices in seconds
  • Train a dedicated model for the closest match
  • Import existing models from a link
  • Use your voice across the whole studio
  • No installs, plugins, or GPUs required
How it works

How to clone a voice

  1. 01

    Add your audio

    Upload up to five clean vocal clips (200mb total), or paste a Hugging Face / Google Drive link to an existing model.

  2. 02

    Pick a method

    Choose from the method picker: Instant clone for a zero-shot voice in seconds, Best · Voice or Best · Acoustics to train a dedicated model, or Use a link to import one.

  3. 03

    We build your clone

    VoiceDub isolates your vocals and either clones instantly or trains a model — Smart training stops when quality peaks.

  4. 04

    Make covers

    Your voice lands in your voices list and in the cover studio, ready to sing, rap, or narrate anything.

Quality vs. speed

Instant clone or trained model

Two ways to clone — pick Instant clone or Best quality straight from the method picker at the top, and switch any time before you create.

Instant clone

  • Zero-shot — ready in seconds
  • Great for quick covers, speech, and singing
  • Works from a single clean clip
  • Lowest credit cost

Best quality

  • Trains a dedicated model for the closest match
  • Cleanest, most consistent output
  • Smart training stops at peak quality
  • Tune max minutes & epochs in Advanced
Use cases

What you can build

From cloning yourself to importing a community model, here’s what voice cloning unlocks.

Clone your own voice

Train on your vocals and sing or narrate anything as yourself.

Recreate a signature tone

Build a reusable model from a handful of clean reference clips.

Import existing models

Bring a .pth (and optional .index) from Hugging Face or Google Drive.

Characters & narration

Make a consistent character or narrator voice for videos and games.

Style

Vocals or acoustics

Best quality comes in two forms in the method picker — Best · Voice and Best · Acoustics — so pick the one that matches your source.

Vocals

Trains a singing or speaking voice from your clips. Keep Isolate vocals on to strip any instrumental so only the voice is learned.

Acoustics

Trains an instrument or sound from your audio instead of a voice — useful for modelling a specific tone, instrument, or texture.

Advanced controls

The training controls, explained

On a Best-quality clone, flip the Simple/Advanced toggle to fine-tune how your model trains. (Instant clone has no settings to tune.)

Isolate vocals
Separates lead vocals from instrumentals so you can train from a full song. Turn off for raw, dry vocals.
Smart training
Stops automatically when quality peaks, so you don’t over- or under-train. Turn off to control limits yourself.
Max minutes
Caps how long training can run when Smart training is off.
Max epochs
Caps how many passes over your data the model can make when Smart training is off.
Import a model

Already have a model? Import it

Choose “Use a link” to bring an existing voice model into VoiceDub without retraining.

Hugging Face

Paste a direct file link to a .pth, or a repo/tree link, and VoiceDub pulls in the model and index.

Google Drive

Paste a shared file link to a .pth (or a .zip/.rar containing the .pth and .index) or a folder link.

Quality

What makes a great clone

The model is only as good as what you feed it — these four things move the needle most.

Clean, dry vocals

No background music or instrumental bleed gives the sharpest clone.

Enough material

30–60 seconds of varied singing or speech covers most of a voice’s range.

Consistent tone

One singer, one mic, similar settings — mixed sources confuse the model.

Right quality mode

Best quality trains a dedicated model when fidelity matters most.

Details

Voice cloning specifications

A quick reference for what cloning supports.

Clone modesInstant clone & Best quality (trained)
Reference clipsUp to 5 files, 200mb total
Per-file limit50mb per clip
Audio formatsmp3, wav, m4a, mp4, mov
Import sourcesHugging Face, Google Drive
Model formats.pth (+ optional .index), .zip / .rar
StylesVocals or Acoustics
Vocal isolationBuilt in — strips instrumentals
Smart trainingAuto-stops at peak quality
Render timeInstant in seconds; trained in minutes
Formats

Supported audio & model formats

Upload common audio or video for training, or import a model file from a link.

MP3WAVM4AMP4MOV.pth model.index file.zip / .rar
After cloning

Put your voice to work

Once a voice is ready it appears in your library and the cover studio. Here’s what you can do with it.

Song covers

Turn any track into a cover sung in your cloned voice.

Link to cover

Paste a YouTube or TikTok link and convert its vocals to your voice.

Text to speech

Type a script and hear it spoken in your cloned voice.

Record & convert

Sing or speak live, then re-voice it instantly in the studio.

Pro tips

Tips for the best clone

  • Use clean, dry vocals with no background music for the sharpest clone.
  • 30–60 seconds of varied singing or speech is plenty for an instant clone.
  • For Best quality, leave Smart training on — it stops when quality peaks.
  • Keep Isolate vocals on when your clips have any instrumental bleed.
  • Pick Best · Acoustics to model an instrument or sound instead of a voice.
  • Add a voice image so your model is easy to spot in the library later.
Why VoiceDub

Why clone with VoiceDub

Everything you need to build and use a custom voice, in one browser tab.

Two ways to clone

Instant zero-shot for speed, or a trained model for the closest match.

Smart training

Training stops automatically when quality peaks — no babysitting epochs.

Built-in isolation

Vocals are separated from instrumentals so you can train from full songs.

Runs in your browser

No installs, downloads, or local GPU — clone from any device.

Compare

Clone a voice vs. use an existing one

Two ways to make a cover on VoiceDub — pick the one that fits your project.

Clone a voice

  • Train a custom model from your samples
  • Make covers in a voice that’s uniquely yours
  • Best for artists and original projects
Clone a voice

Use an existing voice

  • 10,000+ ready-to-use, user-created AI voices
  • No setup — start covering instantly
  • Best for covers, edits, and fan content
Browse voices
Responsible use

Clone responsibly

Voice cloning is powerful — use it with consent.

Get permission

Clone your own voice or voices you’re allowed to use. Don’t clone someone else’s voice without their consent, and don’t use clones to deceive or impersonate.

The toolkit

Everything around your voice

Cloning is one part of the VoiceDub studio. Explore the rest of the toolkit.

FAQ

Voice cloning — frequently asked questions

How many audio clips do I need to clone a voice?

Just one clean clip works for an instant clone; up to five (200mb total) gives a trained model more to learn from.

What’s the difference between Instant clone and Best quality?

Instant clone is zero-shot — it’s ready in seconds and works for speech and singing. Best quality trains a dedicated model, which takes longer but sounds closer to the source.

Can I import a voice model I already have?

Yes. Choose “Use a link” and paste a Hugging Face or Google Drive link to a .pth file (or a .zip/.rar containing the .pth and .index files).

How long does training take?

Instant clones are ready in seconds. Trained models depend on your audio length and settings, but Smart training keeps it as short as possible.

What audio works best for cloning?

Clean, dry vocals with no background music. Keep Isolate vocals on if your clips have any instrumental bleed, and use a consistent single source.

What can I do with a cloned voice?

Once it’s ready it shows up in your voices and the cover studio — turn any song, YouTube/TikTok link, recording, or text into a cover in that voice.

Do I need a powerful computer or any installs?

No. Cloning runs entirely in your browser on any device — there’s nothing to download, install, or configure.

Whose voice am I allowed to clone?

Clone your own voice or voices you have permission to use. Don’t clone someone else’s voice without consent, and follow our Terms.

Ready to clone your voice?

Drop in a few clean clips and build a custom AI voice in minutes — or import a model you already have.