Synthetic voice
A cloned voice belongs to the person who made it.
VoxNotch can clone a voice from a reference recording. That capability deserves a stated position rather than a checkbox, so here is ours.
- Consent
- Required before cloning
- Watermarking
- No removal workflow
- Review
- Human approval before publishing
The position
Owning a recording is not authority to clone the speaker in it
Having the file is not the same as having the right, and the software does not treat them as equivalent.
Your declaration of rights is evidence, not proof
If you tell the system you have permission, that records your representation. It does not establish that the person you are cloning agreed to anything.
Generated speech is watermarked, and we will not help you remove it
Output from the Chatterbox engines carries an inaudible PerTh watermark. No feature intentionally strips it, and we will not add one.
The engine capability record carries the watermark field, and it is served over the API rather than buried in a policy document.
Your voice models stay on your hardware
A cloned voice is a file on your disk. There is no vendor copy, so there is nothing to leak and no account to close.
Consent boundary
Provenance is not consent.
Rights declarations are your representation. They help record intent, but they do not prove the recorded person approved every future use.
For cloned voices, VoxNotch stores provenance: which engine made it, which recording and which speaker it came from, and which voice it descends from. That lineage is useful for audits, but it is not consent by itself.
Why publish this at all
A position you can hold us to is worth more than a checkbox.
Every voice product has a terms-of-service clause saying you must have the rights. Almost none of them say what they think that means, or what they will refuse to build.
These four statements are the ones we are willing to be quoted on, including the fourth one that draws a hard product boundary: no watermark removal, ever, as a matter of design.
Technical notes
- The watermark declaration is a field on the engine capability record in backend/app/services/tts/capabilities.py. The engines that declare it are the Chatterbox ones.
- Watermark durability can vary through mastering, resampling, compression, or video muxing. Treat it as one signal, not the only control in a production voice pipeline.
- Use explicit rights review, human approval, model provenance, and output disclosure together before publishing cloned-voice material.
The rest of the argument.
This page states a position. The privacy page shows the architecture that makes it more than a sentence.