Why brands clone a voice at all
A founder's voice is often the most distinctive asset a small brand owns, and the least scalable. Recording ad reads, phone-system prompts and video narration means booking time, finding quiet and re-recording every time a price or opening hour changes. A good voice clone removes that bottleneck: once the model exists, new audio is a script edit away.
Typical uses are unglamorous and valuable: IVR menus updated the same day your hours change, five lengths of the same social or radio ad, narrated product updates, and sponsor reads for a podcast. Several tools can also carry a cloned voice into other languages, keeping the brand's sound consistent abroad.
The technology is now cheap and quick, which is precisely why the consent and disclosure questions matter more than the tooling.
Consent and ownership: get it in writing
A voice identifies a living person, so voice recordings and the clone built from them involve personal data under UK GDPR. Even when a founder clones their own voice for their own company, put a written agreement in place, because founders sell up, fall out and move on. If you are cloning an employee or a hired voice actor, a signed licence is non-negotiable and performers' rights come into play.
The agreement should cover:
- Permitted uses, listed specifically: advertising, IVR, social content, training material
- Duration, territory and whether the use is exclusive
- Approval rights: does the voice owner sign off every script, or defined categories of use?
- What happens on exit: revocation, deletion of the voice model and any stockpiled generated audio
- Payment terms, especially for staff and professional voice actors
- Where the voice model is stored and who is authorised to generate with it
Need a hand with this?
Our team delivers Podcast Production for UK businesses — with a free initial consultation, transparent fixed quotes and no lock-in contracts. Tell us what you're working on →
Disclosure: what regulators and platforms expect
The ASA's core rule applies unchanged: marketing must not materially mislead. A synthetic voice is not in itself a problem, but implying a real person personally recorded an endorsement they never saw certainly can be. The practical safeguard is simple: the voice owner approves every script before generation, so nothing goes out in their voice that they would not say on tape.
Platform rules add another layer. YouTube, Meta and TikTok all require disclosure of realistic synthetic media in various contexts, and enforcement has tightened year on year, so label AI-generated voice wherever the platform asks. For phone systems, nobody expects an announcement that the menu voice is synthetic, but if a caller asks whether they are speaking to a human, the honest answer must be available. Many brands now add a light note such as "voiced with AI, approved by the founder" and find it builds trust rather than dents it.
Choosing a tool
The market leaders now differ more on governance than raw audio quality. ElevenLabs is the most widely used, offering quick clones from a few minutes of audio and professional-grade clones from longer recordings, with a recorded consent step before cloning is enabled. Resemble AI courts business users with watermarking and tighter enterprise controls. Descript builds cloning into its editing suite, which suits podcasters who want to patch a fluffed line without re-recording. Murf and Play.ht cover the budget end for straightforward voiceover work.
Whichever you pick, check four things: a genuine consent verification process, watermarking or provenance support, clear commercial licensing of the generated audio, and a documented way to delete the voice model when your agreement ends.
The recording checklist for a convincing clone
Clone quality is decided at the microphone, not in the model settings:
- Record in a quiet, soft-furnished room; echo is the number-one clone killer
- Use a decent microphone at a consistent 15 to 20 centimetres
- Provide at least 30 minutes of varied material for a professional clone; short samples produce flat, slightly lifeless results
- Speak naturally at conversational pace; the clone copies delivery as much as tone
- Include range: neutral explanation, upbeat enthusiasm, warm reassurance
- No background music and no aggressive noise reduction; keep the raw WAV files
- Record the consent statement in the same session, in the same voice
Key Takeaway
Never clone a voice without a signed licence covering permitted uses, duration, approval rights and what happens when the person leaves. Record at least 30 minutes of clean, varied audio in a quiet room for a professional-grade clone, and store the consent statement alongside it. Use the clone for IVR, ad variants and routine content; keep apologies, sensitive announcements and live customer conversations for the real human voice.
Where the clone works, and where the real person must show up
Cloned voice is at its best on routine, high-frequency audio: phone menus and out-of-hours messages, ad variants for testing, localised versions of existing content and narration for product updates. These are exactly the jobs that never got recorded properly before, because studio time was too precious for a two-line change.
Keep the real human for anything where sincerity is the point. Apologies, redundancy announcements, crisis responses and personal thank-yous lose all their value the moment anyone suspects they were generated, and suspicion is now the default. A useful test: would discovery of the synthesis embarrass you? If yes, record it live. If you want help planning brand audio, from voice licensing agreements to IVR scripts, our team can help.