Consent-first by design
Built for voices you own or have documented permission to model. Your own voice is the canonical use case, and impersonation is a line, not a feature.
Voice Clone Lab
Train a private model from your own voice — consent first, always — and sing every idea you have.
1
2
3
4
Record at least 15 seconds of your singing (up to 30 seconds), or upload a file. Clean vocals without background music work best.
or
My Voices
Sign in to see your cloned voices.
one voice, many songs
What cloning actually captures
Turn recordings of a voice you have explicit permission to use — usually your own — into a private model that can sing new material on demand.
Training extracts the qualities that make a voice identifiable — timbre, vowel color, transition habits — into a model that can apply them to new melodies and lyrics. It's not a recording being replayed; it's a voice's character, made performable.
The model stays private to your account and works across the workspace: cast it in covers, use it for toplines, keep your releases vocally consistent even when they were written months apart. Its power is exactly why access to it stays yours.
A voice model is an instrument built from a person. Done right, it lets an artist sing a hundred demos without a hundred sessions, keep a consistent vocal identity across projects, and perform in languages they're still learning. Done wrong, it's someone's identity taken without asking — which is why consent isn't a checkbox here; it's the foundation the feature stands on.
The craft side is straightforward: the model learns what you feed it. Consistent, clean, dry recordings of one voice produce a clone that captures its character; noisy samples with music bleed produce a blurry average. Budget your effort for the recording stage and the training rewards it.
Cloning, the responsible way
Quality-driven training
The pipeline rewards clean, consistent samples with a model that keeps the voice's fingerprint — its color, its attack, the way it leaves a note.
Private and reusable
The trained model lives in your account and plugs into the cover and generation tools, turning one recording session into a durable creative asset.
From recordings to a working voice
Three stages: gather with permission, record with discipline, verify with variety. None of them optional.
- 1
Secure the consent
Your voice — proceed. Anyone else's — get written, informed permission covering how the model will be used, and keep it. This is step one for a reason.
- 2
Record for the model
Quiet room, steady mic distance, natural delivery across a range of phrases. Dry vocals only — no music underneath, no effects baked in.
- 3
Test beyond the training set
Try the clone on phrases and registers it never heard. Where it wobbles tells you what additional samples the next training round needs.
Legitimate work for a cloned voice
Artists scaling themselves
Demo new songs in your own voice without booking yourself for every idea, and keep the vocal identity of your catalog coherent.
Multilingual releases
Record your voice once; release in languages you're still mastering, with pronunciation carried by the model and identity carried by you.
Project voice continuity
Studios and story-driven channels keep a character's singing voice consistent across episodes and years — with the voice owner's ongoing agreement.
Rules and craft, in that order
No consent, no clone
A famous voice, a friend's voice, an ex-bandmate's voice — without explicit permission they're all off the table. When permission ends, retire the model.
Consistency over quantity
Ten minutes of uniform, clean recordings outperforms an hour of mixed-quality scraps. One mic, one room, one voice state.
Disclose where it matters
Platforms and collaborators increasingly expect transparency about synthetic vocals. Saying so up front costs nothing; being discovered later costs trust.
Voice cloning questions
Can I clone my own singing voice?
Yes — that's the feature's center of gravity. Record clean samples of yourself, train the model, and your voice becomes available to the cover and generation tools whenever you need it.
How much recording does training need?
Quality dominates duration: several minutes of clean, varied, consistent singing beats hours of rough material. Cover different pitches and phrase shapes so the model learns the voice's full behavior.
Can others use my voice model?
The model is private to your account. Sharing access is your decision — and if you're the voice inside someone else's project, put the terms in writing before training begins.
What happens if consent is withdrawn?
Retire the model and stop publishing new work made with it. Building on a revoked voice isn't just an ethics problem — in a growing number of places it's a legal one.
Make your voice a studio asset
Record it once, train it right, and sing every idea you'll ever have — with permission where it isn't your own.
Start voice cloning