Consent for Voice Cloning

Before cloning a voice, put purpose, term, withdrawal and labelling into a record you can show.

Related

VoicePopular singing and the microphoneSample Clearance

The problem

Cloning a voice takes seconds of audio, but using that clone on a release, in an advertisement or on a platform requires consent you can produce. A verbal yes will not survive a royalty dispute, a takedown, an estate or an appeal. Consent is not a single word of permission. It is a document that states its boundaries, who signed it, what it covers and when it expires.

How it works

A voice is recognisable through timbre, accent, diction habits and breathing, and those are exactly what a clone learns. Consent therefore has to cover likeness, not merely the recording it was trained on. Recording copyright and voice personality rights are separate things. The first governs the master, the second governs whether someone may imitate you. Buying the recording is not buying the imitation. If the training set contains another person's recording, a consent form cannot repair the hole.

Practice

Write at least four things into the agreement. Use: which song, which platform, whether advertising or political content is included. Scope: languages, range, duration, and whether sublicensing is allowed. Term: start and end dates, plus what happens to finished work at expiry. Withdrawal: how it is requested, how many days until takedown, and what happens to material already distributed. A minor needs a guardian's signature. Write it in a language both sides actually read rather than leaving an English template in the folder. Keep the raw dry recording, the model version, generation dates and the operator's name. Keep refusals on file as well; they record what was never approved. Label synthetic output in the credits, metadata and captions, not in a footnote.

What to listen for

In Ich grolle nicht, listen to how declamatory diction carries character. In Schlaf, Kindlein, schlaf, listen to breath and closeness. In Spleen, listen to the French nasal vowels. Diction, breath and accent are what consent must cover, and what a clone flattens most easily and reveals most quickly.

Sources

• 声音人格利益与录音授权的区分，依公开法律评论核对，未抄原文
• Consent practice for synthetic voice, checked against published platform policies, no text copied

Used for fact-checking only; all prose is original.