Voice consent policy

Our principles

A voice is cloned only from recordings the organization has the rights to, and only to dub those recordings. Dubbing must never deceive: it must not put words in someone’s mouth or pass synthetic speech off as their own.

Standard voices by default

Unless a project’s owner chooses otherwise, videos are dubbed with standard voices: synthetic voices whose licences allow this use. They are credited on the licences page.

Cloning: a choice made per project

A project’s owner can choose to clone the speakers’ voices. Each speaker with enough speech in the video then speaks the new language in a voice made from their own lines in that video; speakers with too little speech get a standard voice.

The platform can switch cloning off for an organization. Cloned voices are made only on a GPU server in a European data centre; the voice samples and lines stay on its disk, which is deleted when it stops.

The owner’s confirmation

Before any voice is cloned, the project’s owner confirms once for the project that the organization holds the rights to these recordings and uses the voices only to dub these videos. Who confirmed and when is recorded in the activity log.

The organization remains responsible for having these rights, including informing the people recorded and obtaining their consent where the law requires it.

Voice profiles with recorded consent

A voice that is used outside its own recordings — for example the same voice in several projects — needs a voice profile with the recorded consent of the adult speaker. The consent states:

• who the speaker is, and that they are an adult;
• the use they authorize;
• from when and until when (at most ten years);
• whether the voice may be reused in other projects;
• the evidence of their authorization (for example a signed document or a recording), stored privately.

Expiry and withdrawal

Only the project’s owner sees and manages voice profiles; every change and every download of the evidence is recorded in the activity log.

When a consent expires or is withdrawn, the voice can no longer be used to make new speech. Videos already made are not changed automatically; the owner decides whether to delete them.

What is never allowed

Whatever the project, you must never:

• clone a voice from a public clip, a recording of a celebrity or any media your organization has no rights to;
• use a voice to impersonate someone, to deceive or to commit fraud;
• make a person appear to endorse something or say something they did not say;
• present a dubbed video with cloned voices as the speakers’ own recording in the new language.

Watermarks and honest presentation

Speech made with the Chatterbox model carries an inaudible watermark that marks it as synthetic. Speech made with CosyVoice 3, the default cloning model, carries none.

The file of every dubbed video says, in its title and comment tags, that it was dubbed with Metaphona. When you publish a video with cloned voices, say that its voices were made by AI.

Reporting misuse

If you think your voice, or someone else’s, has been used against this policy, write to [Operator: contact email address]. Tell us where you saw the video and why you think the voice was misused.

We look into every report. While we do, we may suspend the projects or accounts concerned and stop the voice from being used.