Home › Consent-cleared voice cloning
Voice cloning with consent that names the deal
A synthetic voice built on someone who did not fully understand what they signed is a product with a fault line running through it. We scope voice identity work explicitly, pay for it separately, and document it so it holds up.
Cloning is never bundled into a dataset session
The most common shortcut in this industry is a dataset agreement worded broadly enough to permit voice identity replication - “including derivative synthetic works” or similar - so that a corpus session quietly becomes a cloning licence. It is cheap, it is legally shaky, and it is the exact thing that turns into a story.
We keep them separate. A TTS corpus session contributes to a model. A voice cloning engagement builds a recognisable synthetic version of one person’s voice. Different scope, different consent, different fee. Contributors are told plainly which one they are agreeing to, and the page we point them to says the same things this one does.
What a cloning agreement names
- Who. Your organisation, named. Not “a technology client”.
- What for. The application the voice will appear in, described in terms the contributor can picture.
- Where. Territory, rather than a default of worldwide.
- How long. A defined term with a renewal mechanism and a fresh fee.
- Passed to whom. Sublicensing bounded and named, not open-ended.
- Excluded uses. The categories the voice will not be used for, written down.
- Deletion. A documented route, agreed before recording.
Why the strict version is the commercially safer one
Doing this properly costs more per voice. It also gives you a voice you can put in a product, disclose under Article 50 transparency rules, evidence in procurement, and keep using when your customers start asking where it came from. The cheap version gives you a voice you have to hope nobody looks into - and a contributor with a grievance and a public platform.
Recording is delivered to broadcast standard through Flaunt Audio and the 11C Media studio group, with the acoustic consistency a clone needs across the whole capture.
Voice identity questions
What makes voice cloning consent valid?
Specificity. The contributor has to know a synthetic version of their voice will be built, who is building it, what it will be used for, where, and for how long - and be paid for that specifically rather than under a general dataset agreement. Consent buried in broad wording about derivative works is not informed consent and will not read well later.
Can we license a cloned voice in perpetuity?
We would rather structure a long defined term with renewal. Perpetual, irrevocable, worldwide grants are the terms that make good contributors refuse, that attract regulatory attention, and that leave you with no route to renegotiate if the voice becomes central to your product.
Does Article 50 affect deployed synthetic voices?
Article 50 transparency obligations took effect on 2 August 2026 and cover disclosure and marking of synthetic content, including audio. That makes the provenance of the voice a visible question rather than an internal one. See our Article 10 page for the wider position.
What happens if a contributor wants out?
We agree the position before recording, not after. Licences are time-limited, sublicensing is bounded, and deletion routes are documented. Audio already absorbed into a trained model cannot be surgically removed - which is exactly why the protections have to be contractual and agreed up front.
A voice you can name the source of
Scoped, consented and documented before a word is recorded.