Google released two text-to-speech models on September 23, Gemini 3.8 Flash TTS and Flash-Lite TTS, and one feature stands out for working voice actors: they can reproduce a voice from a 30-second sample, but only after the voice's owner records a spoken consent statement that the system checks against the sample. If the two voices do not match, no copy is made.
The two models are aimed at different jobs. Flash TTS is pitched at character work for games, audiobooks, podcasts and interactive media, with line-by-line direction of pacing, accents and cues such as laughs or sighs, and Google claims it holds a voice steady across hours of audio. Flash-Lite targets high-volume dubbing and voice agents. There are over 2,000 stock voices in more than 100 languages and dialects, plus a tool for designing new voices from a written description, and dubbing and localisation companies including HeyGen, Wondercraft and Ollang are among the named partners.
Tech coverage read the launch the same way, with Android Authority describing Gemini as now able to clone a voice and perform scripts like an actor. Every clip the models produce carries Google's SynthID watermark and C2PA content credentials, which Google says are meant to protect voice talent and make content traceable. A footnote adds that voice replication is unavailable in Illinois, Texas, the European Economic Area, the UK, Switzerland and India. Google gives no reason, but the report notes that several of those places have laws treating voiceprints or biometric data as specially protected.
The outlet argues that the stock library, not the clone feature, is the bigger competitive threat to performers, because game characters and narrators are exactly the roles it is designed to fill. It also warns that spoken consent effectively becomes paperwork: a request to read one extra line at the end of a session should be treated like a contract, with questions about purpose, platform, duration and pay, and the phrase allowing clones of a voice "you have the rights to use" puts the focus back on what performers have signed.
It adds that watermarks can help performers too, as one more way to show that a clip circulating in their voice did not come from a session they recorded. The piece connects the launch to California's new rule requiring ads with synthetic narrators to say so. It concludes that the release does not settle how much voice work will move to machines, but shows the largest AI companies now expect to be asked for permission, which makes spoken consent part of a voice actor's job. Its closing advice to performers is simple: know exactly what you are agreeing to before saying it into a microphone.
Sources · 1