Artificial Intelligence
Google releases Gemini 3.8 TTS models
Google announced two Gemini text-to-speech models, said they supported voice creation and replication across more than 100 languages, and outlined availability through Google AI Studio, the Gemini API and Gemini Enterprise.
- Google introduced Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS on 23 September 2026.
- Flash TTS was presented as the model for custom voices and direct performance control, while Flash-Lite was positioned for high-volume, lower-cost use.
- Google said the models supported more than 100 languages and dialects, and that voice replication used a 30-second sample plus consent verification.
- Google said generated audio was watermarked with SynthID and accompanied by C2PA credentials.
- Google said the models were available in Google AI Studio and the Gemini API straight away, with Gemini Enterprise to follow, and it named partners including Figma, HeyGen, Wondercraft, 99.co and Ollang.
Google said it had added two text-to-speech models to the Gemini line, Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS. It described Flash TTS as the more flexible option for custom voices and Flash-Lite as the lower-cost version for heavier use.
The post said Flash TTS could generate new voices from prompts, steer pacing and accent, and work across more than 100 languages and dialects. It also said a voice could be recreated from a 30-second sample if the speaker had given consent, with SynthID watermarks and C2PA credentials attached to the output.
Google said developers could try the models in Google AI Studio and use them through the Gemini API from the same day. It said Gemini Enterprise access would come later, and that Gemini Notebook and Google Vids would also use the models.
The company said Flash TTS took first place on Hume AI's Voice Design Benchmark, while both models ranked first and second on its Overall Quality Index. It also named partners including Agora, LiveKit, Vercel, Figma, HeyGen, Linguana, Wondercraft, 99.co and Ollang.
Named in this story
Companies
- introduced the models and described their rollout
- Hume AI
- the benchmark provider Google cited
- Agora
- a platform Google said could build on the Gemini API
- LiveKit
- a platform Google said could build on the Gemini API
- Vercel
- a platform Google said could build on the Gemini API
- Figma
- one partner Google said was integrating the models
- HeyGen
- one partner Google said was integrating the models
- Linguana
- one partner Google said was integrating the models
- Wondercraft
- one partner Google said was integrating the models
- 99.co
- one partner Google said was integrating the models
- Ollang
- one partner Google said was integrating the models
Organisations
- C2PA
- the provenance standard Google said it used
Products and systems
- Gemini 3.8 Flash TTS
- Google's higher-control text-to-speech model
- Gemini 3.8 Flash-Lite TTS
- Google's lower-cost text-to-speech model
- Gemini Notebook
- one Google product slated to use the models
- Google Vids
- one Google product slated to use the models
- Google AI Studio
- the development tool where the models were available
- Gemini API
- the API through which developers could use the models
- Gemini Enterprise
- the enterprise route Google said would follow
- SynthID
- Google's audio watermarking system
How the source tells it
The piece was a promotional launch note, using superlatives, benchmark claims and repeated safety assurances to present the models as both advanced and responsibly built.
- product hype launch language framed the models as changing voice generation into a creative studio
- scale boosterism repeated emphasis on high-volume use, thousands of voices and support for more than 100 languages
- benchmark boasting comparative rankings on Hume AI tests and blind preference evaluations were used to claim leadership
- reassurance a safety section foregrounded consent checks, watermarking and provenance credentials