Google DeepMind launches Gemini 3.8 Flash TTS voice models
The two new text-to-speech models let developers design custom voices from prompts and direct dialogue line by line.
The facts
- Google DeepMind released Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS, available today in the Gemini API, Google AI Studio, Gemini Notebook and Google Vids, with Gemini Enterprise access coming soon.
- Flash TTS targets creative direction and character design, letting users generate new voices from natural-language prompts and control acting cues, pacing, dialect shifts and backchanneling line by line; Flash-Lite TTS is built for high-volume, cost-efficient use such as dubbing and voice agents.
- Google says the library scales from 30 original voices to more than 2,000 production-ready voices across over 100 languages and dialects, including regional variants like Mexican Spanish, Quebec French and Scots English.
- Voice replication requires a 30-second sample plus a matching verbal consent recording from the voice owner, and every generated clip carries a SynthID watermark and C2PA credentials.
- Google reports Flash TTS ranks #1 overall on Hume AI's Voice Design Benchmark (71.4) and leads accent modelling (60.8), while Flash and Flash-Lite TTS take the #1 and #2 spots on Hume AI's Overall Quality Index.
- Partners integrating the models include Agora, LiveKit, Pipecat, Vercel, Figma, HeyGen, Linguana, Wondercraft, 99.co and Ollang.
Google DeepMind has added two text-to-speech models to the Gemini family: Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS. Both roll out today in the Gemini API and Google AI Studio. Flash TTS also reaches consumers through Gemini Notebook, and Flash-Lite TTS through Google Vids. Enterprise access via Gemini Enterprise is coming soon.
The pitch is a shift from picking a preset voice to directing a performance. Flash TTS is aimed at creators and character designers. It generates original voices from natural-language prompts. Users can script acting cues, dialect shifts, pacing and non-verbal sounds like laughs or sighs, line by line. Flash-Lite TTS trades some of that creative control for throughput, targeting high-volume dubbing and voice agents.
Google says its voice library now runs to more than 2,000 production-ready voices across over 100 languages and dialects, up from 30 original voices. Both models can maintain speaker consistency across hours of continuous audio, which suits podcasts and audiobooks.
Voice replication needs only a 30-second sample. But Google requires a verbal consent recording from the voice's owner that matches the reference speaker before a clone can be created. Every clip carries a SynthID watermark and C2PA credentials. Google frames these as protection for voice talent and a defence against misinformation.
Google reports that Flash TTS ranks first overall on Hume AI's Voice Design Benchmark, scoring 71.4. It also leads on accent modelling with 60.8. Flash and Flash-Lite TTS take the top two spots on Hume AI's Overall Quality Index. In blind preference tests on Voice Arena, Google says the models place highly in languages including Japanese, Brazilian Portuguese, Vietnamese, Modern Standard Arabic, Mexican Spanish and Hindi.
Developer platforms Agora, LiveKit, Pipecat and Vercel are building on the API. Figma, HeyGen, Linguana, Wondercraft, 99.co and Ollang are named as integration partners for dubbing and voice-agent products.
Why it matters
Consent-gated cloning from a 30-second sample, plus mandatory SynthID watermarking, sets a compliance bar for rival voice-cloning tools. Teams building dubbing or voice-agent products should check their current provider's consent flow now. Regulators will likely ask the same question soon.
Primary source
Google DeepMind — deepmind.googleWritten from the primary source above by The Neural Net’s automated desk (claude-sonnet-5, theneuralnet/pipeline@1) without human review. How this works. Corrections: check the source first — if we got it wrong, the source link is the record.