Google’s New Gemini TTS Models Can Clone a Voice from 30 Seconds of Audio
Google (find out more) has released Gemini 3.8 Flash TTS and Flash-Lite TTS. Users can:
- Design new voices from text prompts (over 100 languages and dialects) or choose from over 2,000 ready-made voices.
- Recreate a voice from a 30-second audio sample after consent.
“Today, we’re introducing... two new text-to-speech models,” Google announced.
Both models are rolling out from today in the Gemini API and Google AI Studio. They join the existing Gemini 3.8 family.
Flash TTS vs. Flash-Lite TTS
- Flash TTS: Built for creative work like games, audiobooks, and podcasts. Supports two-speaker scenes, long-form audio, and stage directions.
- Flash-Lite TTS: A cheaper, high-volume option aimed at dubbing and voice agents.
Features:
- Voice customization with regional varieties (e.g., Mexican Spanish, Quebec French).
- Saved custom voices for reuse.
- Upcoming timbre, pitch, pace, and accent adjustments.
- SynthID watermarking and C2PA content credentials on generated clips.
Performance Highlights:
- First place on Hume AI’s Voice Design Benchmark (71.4 score).
- Top positions in blind preference tests on Voice Arena for languages like Japanese, Hindi, and Mexican Spanish.
Competitors include ElevenLabs, Murf AI, and Adobe’s Firefly. New integrations include Figma, HeyGen, Wondercraft, Linguana, 99.co, and Ollang.
Upcoming Releases:
- Flash TTS to Gemini Notebook.
- Flash-Lite TTS to Google Vids.
- Enterprise access through Gemini Enterprise soon.