Synthetic voices and avatars

Synthetic speech has reached the point where short informational narration is broadly acceptable to listeners. Persuasive and emotional delivery is not there.

Where synthetic voice works

The economics are compelling for the top rows. A product video narrated in eight languages costs a fraction of eight voice sessions.

The consent question

Cloning a specific person's voice requires that person's consent, and the consent must be specific.

What it must cover: which uses, in which media, for how long, in which languages, whether it may be edited into new statements, and how the person withdraws consent.

An employee's general employment does not cover it. A voice actor's original session fee does not cover it either unless the contract says so explicitly, and older contracts do not.

Several jurisdictions treat voice as a protected personality attribute, and the direction of legislation is towards stronger protection, not weaker.

Avatars

Synthetic presenters work for explanation and instruction, where the viewer wants information rather than connection.

They fail where the point is a human being: testimonials, founder messages, anything conveying care or trust. The uncanny quality is most noticeable exactly where warmth matters.

The same consent rules apply, more strongly, to any avatar based on a real person's appearance.

Practical rules

1. Never put words in a real person's synthetic mouth without documented consent for that specific statement.
2. Disclose where a voice or presenter is synthetic and could be taken for real.
3. Keep records: whose voice, what consent, which uses, what date.
4. Do not clone a public figure's voice, in any context, including parody, without legal advice.
5. Check the platform terms; several advertising platforms have specific policies on synthetic media.

The regulatory direction

Disclosure obligations for synthetic media are being introduced in multiple jurisdictions, with the clearest requirements around realistic depictions of identifiable people.

Building disclosure and record-keeping into the process now costs almost nothing. Retrofitting it across published material later is expensive.

Where the saving is real

Multilingual versions of existing material is the strongest case. A single well-produced original, translated and narrated synthetically into several languages, delivers reach that would previously not have been funded at all.