Gemini 3.8 Flash-Lite TTS is the faster, lower-cost option, designed to generate large volumes of speech with low latency. Supporting 101 languages, it is intended for high-volume dubbing and audio production, read-aloud features, voice agents, and everyday single-speaker content.
Both models support prebuilt voices, custom voice design, and voice replication. Users can create voices by describing their desired role, accent, and vocal characteristics in natural language. They can also control tone, emotion, pacing, and other aspects of individual lines and add vocal cues such as laughter, sighs, and pauses at specific points in a script. Voice replication requires a 10–30-second reference recording of the user’s own voice, or one they have the rights to use, and a separate consent recording from the speaker.
Fine-grained performance controls were already available in Gemini 3.1 Flash TTS, released in April 2026. That model offered 30 preset voices, supported more than 70 languages, and introduced over 200 audio tags for controlling speech delivery.
The new release adds custom voice design and voice replication. Google also plans to introduce “voice remixing,” which will allow users to modify the timbre, pitch, pace, and accent of voices from its library using natural-language prompts. The feature is not yet available.
A Localization Manager’s Guide to Language AI in Enterprise Software
A guide to Language AI as a Feature (LaaF) in enterprise software, covering AI translation, captions, speech translation, LaaF maturity, and enterprise strategy.
In Hume AI’s independent evaluations, Flash and Flash-Lite ranked first and second, respectively, among 34 models overall, with their largest gains in long-form stability. Flash led the voice-design and accent evaluations, while Flash-Lite ranked seventh among 13 models for voice replication, close to average for speaker similarity. Both models also ranked among the leading systems in Voice Arena’s blind preference tests for Japanese, Brazilian Portuguese, Vietnamese, Modern Standard Arabic, Mexican Spanish, and Hindi.
The models are rolling out through the Gemini API and Google AI Studio. Flash TTS is also available in Gemini Notebook, while Flash-Lite is available in Google Vids. API access through Gemini Enterprise is coming soon. Google said all generated audio is watermarked using SynthID.
The TTS models join Google’s broader Gemini audio lineup, following Gemini 3.5 Live Translate for speech translation, Gemini 3.5 Transcribe for transcription, and Gemini 3.8 Live and Live Extended Thinking for voice interaction. On September 24, Google extended Gemini 3.8 Live with a Live Avatar feature, adding video avatars that can switch across 97 languages and adjust their lip movements and expressions to the language being spoken.