Valid voice cloning consent should tell a future reviewer who authorized the model, which recordings and uses were covered, who may operate it, how scripts are approved, and what happens after withdrawal. A platform confirmation is one control inside that record; it is not the entire permission process.
The consent decision
Before uploading audio, decide whether the speaker understood and authorized both model creation and the intended generated speech. Those are separate decisions. Permission to store an interview recording, for example, does not automatically authorize a reusable synthetic voice.
If either authority is unclear, do not create the model. Return to the rights holder or speaker with a concrete description of the workflow and expected publications.
Method: build an authority record
Create one record that names the speaker, the person granting permission, the relationship between them, the source recordings, permitted project or channels, allowed operators, review process, term, withdrawal route, and retirement owner.
Attach the actual language used to obtain permission. Record whether the speaker could ask questions and whether compensation or other conditions apply. Avoid a vague “all media forever” summary when the discussion was about one course or campaign.
Use representative draft scripts as a comprehension check. Show at least one routine line and one line near the edge of the planned tone or subject. If the second example surprises the speaker, the proposed scope is not yet shared.
Workflow gates
Gate 1: create
Verify the speaker's identity, the grantor's authority, the recording rights, and the written scope. Then use the voice cloning flow only with authorized source material.
Gate 2: generate
Limit model access to the smallest working group. Define which operators may submit scripts and whether new topics require renewed review. Store the private model in My Voice Models rather than circulating credentials or source audio.
Gate 3: publish
Confirm that the final script and medium remain inside scope. Identify the factual approver, apply any agreed disclosure, and retain the accepted file plus script revision in Generation History or the project record.
These gates prevent one broad permission statement from silently standing in for three different acts.
Limits, withdrawal, and retirement
Consent should be reopened when the audience, subject, tone, distribution channel, operator, or commercial relationship changes materially. A harmless training narration does not settle whether the same model may deliver political, medical, financial, or endorsement messaging.
Run a withdrawal drill before launch. The team should be able to identify the model, remove operator access, stop queued publication, locate released files, record the effective date, and explain any copies that cannot be recalled.
At the project's end, choose and document one state: retained under an ongoing grant, disabled pending renewal, or deleted. Do not leave an unused model available merely because nobody owns the retirement step.
Fish Voice cannot determine whether a signature, recording, or grant is legally sufficient for a particular jurisdiction or use. The project owner remains responsible for permission, applicable law, and the Terms of Service. Public imitative styles are also governed by the voice usage policy.
Evidence checklist
- speaker and grantor identity;
- source recording inventory and recording rights;
- explicit permission to create a synthetic model;
- permitted scripts, media, audiences, operators, and duration;
- disclosure and script-approval rules;
- withdrawal contact and response owner;
- final model status after the project.
Review the record at each workflow gate, not only when the model is first created.
FAQ
Is the Fish Voice consent checkpoint enough by itself?
No. It confirms the operator's acknowledgment inside the product. The project still needs evidence of actual authority and scope.
Does owning a recording prove cloning permission?
Not necessarily. Rights in the recording and permission to create or use a synthetic voice can belong to different decisions or parties.
Must every script return to the speaker?
Follow the agreed review rule. Narrow, well-understood uses may use delegated approval; sensitive topics or material scope changes should be escalated.
What should happen when permission is withdrawn?
Execute the documented withdrawal process, restrict further use, preserve the decision record, and address existing publications according to the agreement and applicable obligations.