The line that matters is identifiability. A synthetic singing voice that represents no specific person is a musical instrument. A voice modelled on an identifiable individual is that person's identity, and using it without consent carries genuine legal and ethical exposure in many jurisdictions. The safe route is well established: use generic synthetic voices, or licensed voice models where the singer agreed and is compensated.
Copyright is not usually the operative law here. A voice is not a recording or a composition; it is an attribute of a person.
The frameworks that apply instead vary by country and include rights of publicity or personality, passing off and unfair competition, false endorsement, and in some places recent legislation aimed specifically at digital replicas of voice and likeness. Several of these are new, several are being tested, and they differ enough that a use which is clearly prohibited in one territory may be unaddressed in another.
What they broadly share is the idea that a person has an interest in their own identity being used commercially. A convincing clone engages that interest whatever the technical route to producing it.
It uses them without asking. A singer spends decades developing a voice, and a model can reproduce it from their recordings without any involvement from them.
It creates statements they did not make. A cloned voice can be made to sing anything — words the singer would refuse, endorsements they did not give, material that damages them. This is a harm no licence fee addresses.
It competes with them using themselves. A cloned voice available at no cost competes for the same work the original singer does.
Consent is not a technicality. In every other part of the industry, using someone's performance requires permission. Voice cloning removed the step where they could say no.
Some uses are clearly benign. Demos, private experiments, learning, parody in jurisdictions that protect it, and posthumous work with estate approval are meaningfully different from passing off a fake single as real.
Voice conversion for your own voice. Using a model to change the character of your own singing is a production tool, closer to a vocoder than to impersonation.
Accessibility. Singers who have lost their voice through illness have used the technology to keep working, with consent that is unambiguous because it is their own.
Style is not identity. A vocal that resembles a genre convention rather than a specific person does not implicate anyone.
The disagreement is not really about whether consent matters. It is about where the boundary of identifiability falls, and that boundary is genuinely blurry in the middle.
| Use | Exposure |
|---|
| Generic synthetic voice, no real person modelled | Low |
| Licensed voice model with the singer's agreement | Low, within the licence terms |
| Voice conversion applied to your own performance | Low |
| Clone of an identifiable artist, unreleased and private | Lower, but distribution changes it immediately |
| Clone of an identifiable artist, released | High. Removal, claims, and potential legal action |
| Clone used to imply endorsement | High, and adds false endorsement exposure |
| Clone of a deceased artist | Varies enormously by territory; some protect posthumous rights, some do not |
- Use synthetic voices that model no one. Several providers offer these with clear commercial terms.
- Use licensed voice models. A number of platforms let singers licence their voice and share revenue, which resolves consent and compensation together.
- Get written permission for any voice belonging to a real person, including collaborators. A friend saying it is fine is not a licence.
- Read the licence scope — territories, media, term, whether derivative use is permitted, and whether it covers commercial release.
- Do not imply endorsement. Even with a licence, presenting a track in a way that suggests an artist participated when they did not is a separate problem.
- Disclose where required. Distributor and platform declarations increasingly cover synthetic vocals.
Streaming services and video platforms remove unauthorised voice clones of identifiable artists on request, and this has been one of the most consistently enforced areas of AI policy. Repeat uploads risk account-level consequences.
That enforcement is independent of the legal question. A track can be taken down quickly regardless of whether anyone ever files a claim.
If a listener would think a specific real person sang it, and that person did not agree, do not release it. Everything else in this area is a judgment call; that one is not.
Nothing here is legal advice, and the rules differ substantially by country and are changing. Check the current position in your territory before releasing anything that uses a recognisable voice.
Related reading: training data and music AI ethics, best AI vocal tools, and disclosing AI use in music.
Start making music in Veena
Free, browser-based, no downloads required.
Try Veena Free