Why Cantari Has No Children's Voices

The policy

Cantari offers no child-sounding stock voices, does not permit cloning the voice of a minor, and does not present adult voices as children. This is categorical: there is no parental-permission path, no special tier, no exception. When a child character is genuinely needed in production, the industry answer has always been an adult performer reading young, and every voice we carry supports that.

Reason one: children cannot give the consent cloning depends on

Everything in our cloning feature rests on one legal act: an adult attesting that the voice is theirs or that they have the speaker's permission. That attestation binds because adults can bind themselves.

A minor's agreement does not work that way. In most legal systems a minor's contract can be undone at their option, during childhood or after coming of age. A voice clone built on a child's consent could be legally unwound years later, along with everything made from it. Parental permission exists in law, but verifying it reliably at internet scale, across countries with different age thresholds, for something as permanent as a voice model, is exactly the kind of promise we refuse to make casually.

Reason two: a child's voice is protected data in itself

Under United States children's privacy law, a child's voice recording has counted as protected personal information since 2013, requiring verified parental consent to collect and strict limits on keeping it. Regulators have enforced this against the biggest companies in the world specifically over retained children's voice recordings.

European law goes further: voice data used to identify a person is biometric data, children's data carries heightened protection by design, and consent ages vary by country. A voice clone is, by definition, the long-term retention and productization of a voice recording. For a child's voice, that collides with these laws head on.

Reason three: voiceprint laws with real teeth

Several US states treat a voiceprint as a biometric identifier requiring written notice, a written release, and published retention rules, with fixed money damages per violation and no requirement to prove harm. These laws make no exception for good intentions, and a minor cannot sign the release they require. We would rather not hold the single riskiest category of data a voice company can hold.

Reason four: the law on voice replicas is moving one direction

Tennessee extended its voice and likeness protections to AI replicas in 2024. California added digital-replica consent laws the same year. A federal bill on unauthorized voice replicas keeps advancing, and the EU now requires synthetic audio to be disclosed as synthetic and outright prohibits AI that exploits children's vulnerabilities. Every new law in this area points the same way: stricter, with children treated as the people most worth protecting. Excluding children's voices means none of these laws will ever surprise us.

Reason five: the documented abuse record

This is the reason that would be sufficient alone. Cloned children's voices have already been used in staged kidnapping ransom calls to parents, a case serious enough to reach congressional testimony. Child-safety organizations document rising AI-generated child exploitation material, including synthetic audio. A child-sounding voice on a self-serve platform would be a tool whose legitimate uses are easily served other ways, and whose illegitimate uses are exactly these. No safeguard we could engineer changes that arithmetic.

What this means in the product

If you believe a voice on Cantari was cloned from a minor, report it from the voice or clip page, or write to us directly: it goes to the front of the review queue.

• The voice roster contains adult voices only, and always will.
• The cloning consent step is an adult's attestation; cloning a minor's voice breaches the terms and such voices are removed when found.
• Community content that sexualizes or endangers children is blocked at the severe tier of moderation, never merely held for review.
• We do not guess ages from audio: that inference is unreliable and is itself a biometric judgment we decline to make. Enforcement works through policy, attestation, reporting, and human review.