Signature voice

A signature voice is a synthetic voice built from recordings of a real person, typically a journalist or a recognisable figure of the newsroom, used to narrate the publication's articles. Where a generic neural voice is shared across many sites, a signature voice belongs to one media outlet alone: it becomes part of the sound identity, like a layout or a logo.

How a voice is cloned

The principle: the person records a speech sample, from which a model learns the timbre, pace and pronunciation habits specific to that voice. The model can then read any text in that voice, including articles the person never read aloud. Quality depends on the sample: a clean recording, free of background noise, in a register close to article narration, gives the best results.

The legal framework is not optional

This is what sets the signature voice apart from every other audio building block. Under European law, a voice can identify a person: it is personal data, and when processed to uniquely identify someone it falls under biometric data within the meaning of the GDPR, a specially protected category.

In practice, cloning a journalist's voice requires explicit written consent, a contract specifying the authorised uses (which content, for how long), and a real ability to withdraw that consent, with everything that implies: stopping use and deleting the model. Departures must be anticipated too: what happens to the voice of a journalist who leaves the newsroom? A serious publisher settles these questions in writing before the first recording session, not after. WeDispatch's commitments on model storage and deletion are detailed on the security page.

Why go to the trouble

Because the voice carries the relationship. A voice readers already know extends their trust into audio, and it cannot be copied by a competitor the way a generic voice can. The signature voice feature manages the process end to end, from recording to deployment across articles.