Can a licensed synthetic TTS voice be used as reference audio for Gemini 3.8 Flash TTS Voice Replication?

Gemini API
1

Hi Google AI team,

I have a question specifically about the new Voice Replication feature of Gemini 3.8 Flash TTS (gemini-3.8-flash-tts) using the /v1beta/voices API.

I’m developing a commercial Japanese-learning application and currently use the synthetic voice model “Mai (まい)” from AivisSpeech.

Mai is distributed under the Aivis Common Model License (ACML) 1.0. The license permits commercial use and derivative works subject to its terms.

What I would like to do is preserve Mai’s existing timbre while benefiting from Gemini 3.8 Flash TTS’s higher-quality speech generation, pronunciation control, and natural prosody.

The proposed workflow would be:

  1. Generate a 10–30 second reference clip using the licensed Mai TTS model.

  2. Generate the required consent phrase using the same Mai TTS model, so both clips have exactly the same synthetic voice identity.

  3. Submit those clips to the Gemini API /v1beta/voices Voice Replication endpoint.

  4. Use the returned replicated voice_... ID with gemini-3.8-flash-tts.

  5. Use the resulting generated speech in our commercial Japanese-learning application.

I understand that the current Voice Replication documentation describes the reference and consent recordings in terms of a human speaker.

My question is about the case where the source voice is already a licensed synthetic voice model, rather than an attempt to clone an individual person’s voice.

Could you please clarify:

  1. Is reference audio generated by a third-party synthetic TTS model allowed as input to Gemini 3.8 Voice Replication when the user has appropriate rights to use and create derivative works from that voice model?

  2. If the reference voice is synthetic, may the required consent phrase also be generated by that same synthetic voice model?

  3. Does a license such as ACML 1.0 that permits commercial use and derivative works satisfy Google’s authorization requirement for this use case, or does Google still require separate consent from the human speaker whose recordings may originally have been used to train the third-party TTS model?

  4. If synthetic reference audio is intentionally not supported by the /v1beta/voices Voice Replication feature, is there another supported Gemini 3.8 TTS workflow for preserving the timbre of an appropriately licensed synthetic voice?

I’m not trying to replicate an identifiable person’s voice without authorization. The goal is specifically to migrate or enhance an already licensed synthetic TTS voice while complying with both the source model license and Google’s Voice Replication policy.

I’d like to confirm the policy before testing or using this workflow in production.

Thank you.

3

Hi Google AI team,

We use two MiniMax preset synthetic voices for two fictional characters. We want to migrate to Gemini TTS while preserving each character’s existing vocal identity; a newly designed similar voice would not meet our requirement.

The Voice Replication documentation requires two real human recordings from the same adult speaker: reference audio and consent audio. We do not intend to replace human consent verification with a synthesized consent statement.

Could you clarify:

  1. Is there an officially supported migration or registration route, or an individual review process, for third-party synthetic voices where appropriate rights to replication can be established?
  2. What specific permissions from the source provider or underlying speaker would be required? We have not established that ordinary use of a preset voice grants cross-provider replication rights.
  3. If this is unsupported, please confirm whether any other Google TTS product supports this use case without changing the character’s voice.

We will not submit audio or create replicated voices until the supported requirements are clear. No audio, voice IDs, API keys, or personal information are included in this question.

Thank you.