The Neuron reports Gemini 3.8 Live supports speech-to-speech interaction across 97 languages with simultaneous vision and audio input
Concluded·Day 10·in the edition since 16 Sep 2026·4 sources·updated 25 Sep 2026
Google DeepMind's Gemini 3.8 Flash TTS
The gist
Google DeepMind Launches Gemini 3.8 Flash TTS Voice Models
The models offer voice design from text prompts and voice cloning from short audio samples at low per-use cost. Built-in safety measures, including SynthID watermarking on all generated audio, are intended to keep AI-generated speech detectable.
The full picture
Google DeepMind launched Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS on September 23, 2026, making both available through the Gemini API and Google AI Studio. Flash TTS supports generating new voices from text prompts, selecting from over 2,000 production-ready voices, or replicating a voice from a 30-second audio sample, with multilingual support across more than 100 languages. Flash-Lite TTS targets high-volume, cost-efficient use cases such as dubbing and voice agents. Google DeepMind stated that Flash TTS secured the #1 overall spot on Hume AI's Voice Design Benchmark with a score of 71.4, and both models took the top two spots on Hume AI's Overall Quality Index. Safety measures include consent verification for voice replication, SynthID watermarking on all generated audio, and C2PA credentials.
How it developed
The Neuron on September 24 corroborated Gemini 3.8 Flash TTS's core features, confirming promptable voice design, a library of over 2,000 stock voices, 30-second consented voice replication, and multilingual support across more than 100 languages.
Google DeepMind had released Flash TTS alongside Flash-Lite TTS, a companion model targeting high-volume applications such as dubbing and voice agents, through the Gemini API on September 23.
First citedThe Neuron
Simon Willison published a hands-on playground on September 24 for Gemini 3.8 Flash TTS, the voice model Google DeepMind released the previous day through the Gemini API, reporting that generating 1 minute 18 seconds of audio took about 20 seconds and cost 2.74 cents.
The playground also demonstrated the API's multi-character conversation support, where each character can carry distinct voices and style instructions.
Sources
DeepMind Blog
Google DeepMind Twitter
Simon Willison
Related
- Continues fromGoogle DeepMind Gemini 3.8 Live voice models
- Merged inGoogle DeepMind's Gemini Live Avatar
The daily email
Want this in your inbox?
I send one email each morning with the stories that moved. If you would rather just read here, that works too.
Subscribe free