01
What Google means by voice replication
Gemini 3.8 Flash TTS can create a reusable vocal profile from about 30 seconds of a speaker's reference audio. Google calls the feature voice replication and separates it from prompt-based voice design, which creates a new synthetic voice without copying a real speaker.
A short sample is a product input, not proof of permission. Google's documented workflow requires the user to own the voice or have rights to use it before a replicated profile is created.
02
The separate verbal-consent step
Google requires a verbal consent recording from the voice owner and verifies that the consent speaker matches the person in the reference audio. This makes the launch materially different from an unrestricted upload-and-clone workflow.
Production teams should preserve evidence of rights and consent independently of the technical profile. API support does not remove contractual, platform or disclosure obligations around synthetic speech.
03
Where voice replication is unavailable
At launch, Google says AI Studio voice replication is unavailable in Illinois, Texas, the European Economic Area, the United Kingdom, Switzerland and India. That list is time-sensitive and should be checked before deployment.
The restriction should not be generalized to every Gemini TTS capability. A user may have access to speech generation while the replication workflow itself remains unavailable in the relevant region or surface.
04
SynthID, C2PA and publishing safeguards
Google says Gemini Audio outputs contain imperceptible SynthID watermarking, and replicated voices additionally use C2PA credentials. These provide provenance signals but do not substitute for permission or appropriate disclosure.
For fictional narrators or brand characters, a newly designed synthetic voice may avoid some identity-governance complexity. For a real replicated speaker, test similarity, multilingual behavior and long-form consistency only with consented material.
05
A practical, consent-first workflow
Decide first whether a recognizable real voice is necessary, obtain rights, record a clean reference, complete the matching consent step, create the profile on a supported surface and test it on representative scripts.
Regional availability, model access, quotas and pricing can change. Voice Remix was announced as coming soon and should not be described as part of the currently verified replication workflow until Google documents its release.
Sources
Primary and supporting sources
Use the linked documentation to check current pricing, access and product behaviour. The dates above describe this article; they do not guarantee that every vendor claim remains current.