Cloning a voice

Feed a tool a minute of someone talking and it can produce brand-new speech in their voice — words they never actually said. Useful, unsettling, and easy to misuse. How does a voice get copied, and where's the line?

A voice from a sample

Text-to-speech gives you a menu of generic voices. Voice cloning goes further: give the tool a short recording of one particular person — a minute or two is often enough — and it learns the fingerprint of that voice, its pitch, accent and rhythm. After that, you can type any text and hear it in their voice.

Sample in, matched voice out

It's genuinely handy in the right hands: an audiobook narrator scales their voice across a whole series, a creator dubs their own videos, someone losing their voice to illness banks it while they still can. In the scene below you've got a reference sample — add consent, then clone, and watch a new voice take on the shape of the original.

Consent is not optional

Here's the hard rule, and it isn't a technicality. Cloning someone's voice without their clear permission is how impersonation, fraud and fake 'endorsements' happen. Reputable tools make you confirm you have the right to a voice before they'll clone it, and many add an inaudible watermark. The technology is neutral; using it on someone who didn't agree is not.

The gist

• —Voice cloning recreates one specific person's voice from a short sample, then speaks any new text in it
• —Great uses exist — narration at scale, self-dubbing, saving a voice before illness takes it
• —Consent is the hard rule: never clone a real person's voice without their clear, informed permission