Google releases two Gemini text-to-speech models with voice design

By Splitfeed EditorialLast updated Sep 26, 2026, 19:17 UTC

On September 23, 2026, Google announced two new text-to-speech models, Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS. Both support more than 100 languages, allow per-line stage directions to shape delivery, and can generate two-voice dialogue from a single script. The Flash TTS model goes further by letting users create entirely new voices from plain text descriptions, a feature absent from the lighter Flash-Lite version. Both models include a voice cloning capability that builds a voice profile from a 30-second audio sample. Google released no pricing, availability dates, or access details alongside the announcement.

The combination of features marks a meaningful step beyond conventional text-to-speech tools. Generating a voice from a text description, rather than recording or selecting from a preset library, moves voice creation into a design space where no audio input is required at all. The 30-second cloning feature further lowers the barrier to reproducing a real person's voice. Taken together, these capabilities widen access to synthetic voice production considerably, which is precisely what puts the announcement at the center of an ongoing argument about creative opportunity versus misuse potential. What remains genuinely unresolved is how Google intends to govern cloning and synthetic voice design, since the announcement included no details on safeguards, consent mechanisms, or usage restrictions.

The Pro-AI camp is framing the release in enthusiastic terms. GadgetBond.com described the new models as turning AI voice generation into something closer to a virtual voice studio, emphasizing the creative and practical reach of the toolset. Marie Haynes highlighted the voice replication feature directly, noting that users can now replicate their own voice in Google AI Studio, which signals perceived value for personal and professional workflows. The Anti-AI camp has not yet published reactions to this specific release, though stories involving voice cloning and synthetic voice creation from minimal audio samples typically draw concern from that camp around consent, impersonation risk, and the potential for audio deepfakes. The Middle Ground camp has also not weighed in yet, but stories of this kind usually prompt that camp to acknowledge the creative utility while calling for clearer platform-level rules before wide deployment.

The details most worth watching are any governance or policy terms Google attaches to access when it does announce availability, along with whether the voice cloning feature includes any verification or consent layer for the identity being cloned.