Voice safety / Evidence guide

Safe and authorized AI voice creation

A convincing voice can affect identity, privacy and trust. The safest workflow begins with authorization and purpose, limits access to the account owner, reviews the finished context and clearly discloses synthetic speech when listeners could otherwise be misled.

Published: 2026-08-26
Reviewed: 2026-08-26
Reading time: 7 minutes
Author: Donatello Editorial

Key takeaway

Use a built-in or fully synthetic voice when you do not need a real person's identity. Use a private voice only with documented authorization, never to impersonate or deceive, and review the final audio in context before publishing.

01 / Voice safety

Choose the least identity-sensitive method

Donatello Voice offers built-in text-to-speech and private synthetic Voice Design. Creating new sample-based voice clones and instant references is currently paused; private voices already saved remain accessible only to their account owner.

Need	Recommended path	Identity risk
General narration	Built-in voice	Low when the script is lawful and not deceptive
Distinct original character	Synthetic Voice Design	Lower because it does not require a real-person sample
A specific real person's voice	Documented authorization and a permitted private workflow	High; consent and context must be verified

02 / Voice safety

Document authorization before processing

Authorization should identify the person, the intended project, where the output may appear, whether it may be monetized and when permission ends. A message saying only ‘you can use my recording’ may not cover a synthetic replica used in advertising or future projects.

Confirm the speaker is an adult able to grant permission.
Describe synthetic processing in plain language.
Record the allowed channels, territory, duration and commercial scope.
Provide a way to withdraw future use where the agreement allows it.
Never use scraped calls, social videos or leaked recordings as samples.

03 / Voice safety

Consent to a voice is not consent to every script

Review each script and final edit against the authorized purpose. A legitimate sample does not make a fraudulent endorsement, political statement, medical claim or humiliating context acceptable.

Do not use synthetic audio to impersonate, deceive, defraud, harass or conceal the AI-generated nature of content where that omission would mislead a reasonable listener.

Read Donatello's Acceptable Use terms U.S. Copyright Office report on digital replicas

04 / Voice safety

A safe production checklist

Treat authorization, generation and publication as separate gates. A project moves forward only when all three are clear.

Use a private account and do not share private voice access.
Keep prompts and scripts free of sensitive personal information.
Listen to the complete output for unintended words or emotional meaning.
Label synthetic audio when the context could imply a real recording.
Download the approved file before the 72-hour storage period ends.
Remove or stop distributing a result if authorization no longer covers it.

Verification

Sources and current facts

Product conditions can change. Donatello's dated Product Facts and binding Terms take precedence over this explanatory article if a material detail is updated.