Google launches Gemini 3.8 Flash TTS, Flash‑Lite TTS and Live Avatar

New voice models and Live Avatar aim to make Gemini’s spoken interactions more natural, multilingual and traceable.

Updated 26 September 2026

Google has rolled out two Gemini 3.8 text‑to‑speech models and a new Live Avatar feature, all aimed at richer, more controllable multimodal experiences for apps and enterprise agents. Gemini 3.8 Flash TTS focuses on highly directed, character‑level voice design, including prompt‑based custom voices, replication from a 30‑second reference sample, and control over cues like pacing, dialect shifts, and backchanneling, while Flash‑Lite TTS targets cheaper, high‑volume use cases. Both TTS models support over 100 languages, ship with a library of 2,000‑plus ready‑made voices, and rank at or near the top of Hume AI’s Voice Design Benchmark and Overall Quality Index, according to Google. In parallel, Gemini 3.8 Live with Live Avatar lets Gemini Enterprise customers talk to an animated persona that lip‑syncs and emotes in near real time, including mid‑conversation language switching across 97 languages and asynchronous tool calls that run while the avatar keeps talking. All these audio and avatar outputs are watermarked with SynthID and gated by consent checks for voice cloning, which matters if you’re integrating them into products that need traceability, brand safety, and identity protection at scale.