Cloning a voice — designing the consent

What the consent must state

• Who provided the voice
• The specific permitted use — not "internal use"
• A duration, not indefinite
• Whether it may be shared with third parties
• What happens if they withdraw consent
• What happens when they leave the organisation

Be specific about use

A voice contributed for training-video narration later answering sales calls is outside what the person agreed to, whatever the paperwork says. Add uses by taking additional consent; asking for blanket consent up front raises the refusal rate.

Build the withdrawal path

Without a route to say stop, the original consent was not freely given. Decide in advance what happens to already-published material — typically new generation halts and published assets remain — and agree that in writing.

When to disclose

Internal training rarely needs it. Customer-facing conversation, especially voice response, should not create a situation where someone believes they are speaking to a person. The test is whether they would feel misled on finding out.

Do not clone other people

Imitating a public figure or a professional voice actor carries rights and reputational exposure regardless of technical feasibility. Restrict yourself to synthetic library voices and consenting employees.

Storage and deletion

The voice model is personal data. Limit who can access it and add "delete the voice model" to your offboarding checklist.

FAQ