I have a question about the permitted use of audio generated by Gemini TTS.
I am currently using `gemini-2.5-flash-preview-tts` with the Erinome preset voice, and I understand that this is currently a preview model.
I am creating a narrated picture book and I like the voice quality very much. However, in my use case, I sometimes notice variations in speaker tone when generating the narration sentence by sentence.
As a workaround, I am experimenting with Irodori-TTS, an open-source TTS model that supports speaker inversion.
My current experimental workflow is:
1. Generate several short audio samples with `gemini-2.5-flash-preview-tts` using the Erinome preset voice.
2. Use those generated audio samples as reference/training material for Irodori-TTS speaker inversion.
3. Keep the Irodori-TTS base model frozen and train only the speaker embedding / speaker tokens.
4. Use that speaker embedding to generate the narration sentence by sentence with a more consistent speaker tone.
I do not plan to distribute the speaker embedding itself.
If this method works well, I may use the final generated narration audio in a picture book for commercial or non-commercial distribution.
My question is:
Would using audio generated by `gemini-2.5-flash-preview-tts` in this way for speaker inversion in another TTS model violate the Gemini API / Google AI Studio terms, including the restriction related to developing models that compete with the Services?
This is currently only a technical experiment, and I would like to confirm the terms before using the resulting audio in a published work.
Thank you for your help.