Google released Voice Remixing for Gemini 3.8 TTS on November 4. It lets developers start with a prebuilt or previously designed voice, describe a change in natural language and save the result as a separate voice.
Start with a voice, then change its delivery
Instead of building every voice from scratch, a creator can ask for a change in pace or accent. Google’s release notes describe the result as a new stored voice with its own identifier. The starting voice remains available.
That separation makes comparisons practical. Use the same short script for each version and listen for one difference at a time: pace, emphasis or intelligibility. A waveform alone cannot tell you whether the result sounds right for a particular audience.
Where the feature fits
Google says remixing is available interactively in AI Studio and through the Gemini API. The API creates a voice with Gemini 3.8 Flash TTS; the resulting voice ID can then be used to generate speech with Flash TTS or Flash-Lite TTS.
Remixing is distinct from designing a voice from a description or replicating a person’s voice from a recording. Check the available base voices and access terms for your project before starting; permission to use a personal recording should never be assumed.
A small listening test
Keep the words constant and change one instruction at a time. Save the original, compare the new versions and choose the one that serves the script. This makes it easier to hear which instruction caused a useful change.
