Google makes TTS 3.8 available and requires recording consent to replicate voices. See retention, biometrics and enterprise governance.
Direct answer
Google announced on September 22, 2026 the general availability of Gemini 3.8 Flash TTS and Flash-Lite TTS models and opened the Voices endpoint for creating and managing voices. In replication, the API requires a reference sample and a consent recording from the same adult. Persistent profiles remain in the project for up to a year; server-side persistence mode returns a customer-managed key with a seven-day validity. Companies must treat voice as biometric data, limit purpose, access, retention and revocation and clearly signal when audio is synthetic.
Technical consent does not replace legal basis
Mandatory recording proves a speaker's statement, but the organization still needs to define purpose, transparency, rights and rules applicable in each market.
Persistence changes the risk profile
A reusable tag simplifies production but expands exposure. Sensitive projects may prefer short-duration switches and their own controls, evaluating the operational cost.
Revocation needs to reach the product
Deleting the profile in the API does not automatically remove derived files. An inventory of audios, versions and destinations is necessary to withdraw consent.
Quality must be tested by context
Accent, proper names, emotion and long sessions vary. Acceptance must compare intelligibility, consistency and suitability by language and channel.
Nexus Reading
Before scaling, register the holder, purpose, storage method, term and authorized uses; mark synthetic exits and maintain a verifiable exclusion route.
FAQ
Does the API replicate a voice without recorded consent?
Documentation requires reference audio and consent audio from the same adult speaker for each creation.
What is the difference between storage modes?
Persistent mode saves the profile in the project; non-persistence mode returns a customer-managed, short-lived encryption key.
Can a replicated voice be used for any purpose?
No. Consent, contract, legislation, vendor policy and stated purpose continue to limit use.
Essential guides to delve deeper into the decision
This editorial analysis was produced by Nexus from the official sources below, consulted on September 28, 2026. The text is original and interprets practical implications for companies.
- Google AI for Developers — Gemini API release notes: GA release of Gemini 3.8 TTS models, Voices API, and voice design and replication capabilities.
- Google AI for Developers — Voice replication: required samples, consent, storage modes, limits, retention and management of voices.
Date reported by the main source: September 22, 2026.