Voice replication on gemini-3.8-flash-tts: EU availability, and re-creating stateless voice keys

We’re on a paid Gemini API tier, as a business established in the EU. We’d like to offer voice replication with gemini-3.8-flash-tts to our end users, who record the consent statement themselves. We couldn’t find answers to two questions in the docs:

  1. Is voice replication available to a paid EU customer whose end users are worldwide, or is there an allow-list?
  2. In stateless mode (store=False), the voicekey_ expires after about 7 days. May we create a new key from the same source and consent recordings we stored (they’d pass your verification again), or does the speaker have to record again?

If someone on the team knows: is SynthID embedded in TTS output, and does voices.delete remove the stored consent and source audio?

Hello @Kivanc_Yuksel ,

Thank you for the detailed questions. We can’t provide legal guidance on this forum about EU regulatory compliance or about storing and re-using your end users’ consent recordings. For those points, please review the Gemini API Additional Terms of Service (including the Paid Services section, which applies when you offer your app to users in the European Economic Area, Switzerland, or the United Kingdom) and the Available regions page.

On the technical side:

  • The Voice replication docs do not list a separate allow-list for creating replicated voices through the Voices API.
  • Each voices.create request for a replicated voice takes a reference recording and a consent recording, and verifies them on every call. In stateless mode the returned voicekey_… key is valid for 7 days, while a stored voice (voice_…) is kept for 1 year from its last use (Custom voice limits and TTL).
  • In the Voices API reference, the reference and consent audio are input-only and never returned in responses, and voices.delete deletes a stored voice from your project. For how data is handled on paid projects, please see the Paid Services section of the terms linked above.

Adding to this questions:
are there any plans to offer gemini-3.8-flash-tts from a EU (multi-region or specific location) model endpoint?