On November 4, Google added Voice remixing to the Gemini API changelog: describe a change in one sentence, and an existing voice becomes a new timbre, saved as an independent voice_ ID for repeated use. The feature targets Gemini 3.8 Flash TTS and Flash-Lite TTS, and can also be used interactively in Google AI Studio.
Change a voice in one sentence — here's how
Start from a prebuilt voice, the Extended Voice Library, or a saved custom voice, and describe the change in natural language — a different accent, a different pitch, a slower pace, an older sound — and a call to POST /v1beta/voices returns a new, independent voice to synthesize speech with. A remixed voice can itself be remixed again, so a signature voice can be iterated layer by layer.
It is not voice cloning
Voice remixing changes an existing voice; it does not conjure a clone of a real person out of nothing. It splits the work differently from Voice design and Voice replication, launched in September: the former describes a brand-new voice from scratch in words, the latter replicates a real voice with consent. The September 22 changelog shows that Gemini 3.8 Flash TTS and Flash-Lite TTS became generally available, with the companion Voices endpoint launching the same day — Voice remixing is built on that endpoint.
For podcasts, courses, audio content, and voice assistants, this turns "tuning a voice" from a settings panel into a single sentence, so a fixed character voice no longer depends on fumbling through parameters by trial and error. But the easier voices are to change, the more the licensing boundary matters — whose voice a remix resembles, and whether it can be used commercially, are things creators still need to confirm before using it. When choosing a tool, it is also worth checking how to choose an AI voice cloning tool, compared across podcasts, audiobooks, and short video.
