Generated voices have crept in everywhere: marketplaces, sound libraries, demos received by email. Since June 2026, the French Court of Cassation protects the voice like the image: a person only needs to recognize themselves to act, and it's your campaign they target, not your vendor. Broadcasting a voice whose origin you don't know is a real legal risk, on top of an authenticity your audience senses without being able to explain it. An AI voice detector helps, on one condition: knowing what to listen for.
A rhythm that's too regular, as if metronomed. Breaths that are missing, or placed in the wrong spots. An intonation that falls the same way at the end of every sentence. Approximate French liaisons. And over time, an absence of intent: the voice reads, it doesn't tell.
The best models fool a quick listen. So a voice's origin is no longer verified by ear alone: it's documented. Who recorded, where, under which contract, with which rights assigned. A written chain of rights protects the brand the day the question comes up. And the transparency obligation is no longer on the horizon: since 2 August 2026, whoever broadcasts must state that content is AI-generated, with no delay and no exception for the advertiser.
Ask for the origin in writing: who recorded, when, under which contract. Read your supplier's warranty clause, that's what says who bears the risk, not the product page. And if the voice is indeed generated, label it: in France, the ARPP recommends the wording "AI-generated voice". Read more: what the law really says.
Paste a script into VoiceVerify: the tool generates the synthetic version, you listen to it, and the analysis details what a human performance would bring to this exact text. Free.
At FRVOICEOVER, every voice we deliver is human, recorded in our studio, with a documented chain of rights. It's been our job for 15 years, from dubbing to commercials. Tell us about your project.