Voice synthesis applied to real, identifiable persons is not neutral. The same technique used for ADR restoration can be used for fraud. Context and consent are what separate documentary practice from deepfake abuse.
This framework applies when using this toolkit on real persons' voices.
1. Informed consent
The subject must understand:
- That their voice is being digitally cloned
- What that clone will be used for (specific film/project)
- That the synthesis may produce speech they did not record
- Their approval rights on released material
Minimum: written consent specifying the production, the use case, and approval process.
2. Declared artistic frame
The film/work in which the voice synthesis appears must declare its use of AI synthesis, either in the work itself or in its credits/press notes. Voice synthesis presented as authentic unaltered recording is a different category.
3. Scope limitation
Voice model and generated audio are for the specified production only. No secondary use, licensing, or transfer to third parties without new consent.
VOICE SYNTHESIS CONSENT
Subject: [Name]
Production: [Film title, director, production company]
Date: [Date]
I consent to:
- Recording of my voice for digital voice model training
- Creation of a voice synthesis model from these recordings
- Use of this model within the production [title] to generate speech
I retain:
- Right to review and approve synthesized speech before public release
- Right to withdraw consent for specific uses that fall outside the agreed frame
Signed: _______________ Date: _______________
Post-mortem voice reconstruction: Use of archival recordings to reconstruct a deceased person's voice. Requires estate consent and is subject to local law. Historical documentation framing (e.g., "this is what they would have sounded like") is stronger than presenting synthesis as authentic recording.
Reconstruction of things actually said: Using synthesis to reconstruct a line that was recorded but technically unusable (noise, interruption). Lower ethical weight than generating new speech, closer to ADR. Still requires consent.
Non-consenting subjects: No use case justifies voice synthesis of a person who has explicitly refused, or who cannot give informed consent.
This toolkit can be misused. Publishing it with this ethical framework is not a guarantee of ethical use. It is a record of the conditions under which the authors use these techniques.
If you are unsure whether your use case is ethical: ask the subject. That's the test.