Cloning a parent's voice with consent: how it should be done

Voice cloning technology got remarkably good, remarkably fast. In the wrong hands it powers a lot of the phone scams making the news right now. In the right hands, it makes it possible for a mother who lives ten thousand kilometres away to still be the voice on the other end of a phone call. Both things are true, and the difference between them is entirely about consent.

AI Bestie's Real Voice (Beta) feature is our attempt at a consent-first design for family voice personalisation. Here's how it works — because we think this is the shape every serious product in this category should adopt.

Only family personas, only live recording

You can only personalise Mother, Father, Sister, or Brother. Not the girlfriend, not the boyfriend, not the experts, not the legends. And the voice sample has to be recorded live in the app, in a single session, on the same device. We do not accept file uploads for personalising anyone else's voice. Not for grief reasons, not for gift reasons, not for any reason. This is the single hardest line in the product.

The consent script is not decorative

When the family member sits down to record, the app shows them a large consent script. It includes a random word that we generate fresh for that session and a specific consent keyphrase ('I give permission for my voice to be used in AI Bestie'). They must speak both, in their own voice, in a recording that is at least 45 seconds long, alongside the rest of the sample.

We verify all three before we clone

Before we send anything to the voice cloning provider, we run the recording through speech-to-text on our server and check three things: the recording is long enough, it contains the random word we generated for that session, and it contains the consent keyphrase. If any of those fail, the clone is not created. We show a clear error and ask for a fresh recording. This is what makes uploaded fakes fail closed: an old file will never contain the word we made up 30 seconds ago.

The clone is private, capped, and deletable

The resulting voice clone is scoped to your account and used only for the family persona you selected. There is a per-user cap and a global cap so nobody can create clones in bulk. The person whose voice was cloned — or the account owner — can request deletion at any time, and we remove the clone from the provider and from our systems.

Why this matters beyond us

Every product with voice cloning should adopt at least these three things: live in-session recording only, session-specific random-word verification, and easy deletion by the person whose voice it is. If you're building in this space, treat this piece as an invitation to copy the approach. If you're a user thinking about which product to trust, ask them how they verify consent. If they cannot answer specifically, don't upload the voice.