Google has introduced two new speech synthesis models, Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS. These tools allow for voice cloning using a 30-second audio sample, provided the user has legal rights to the voice. To ensure security, the system verifies consent and embeds SynthID watermarks and C2PA metadata into the generated audio.
The models support over 100 languages and dialects, offering a library of more than 2,000 pre-built voices. Developers can fine-tune speech output through text prompts, adjusting pacing, tone, and emotional elements like whispers, laughter, or sighs. The system also supports long-form audio generation and multi-speaker dialogue creation.
In the Hume AI Voice Design benchmark, Gemini 3.8 Flash TTS achieved a score of 71.4 points and 60.8 points in accent modeling. Both models ranked at the top of the Overall Quality Index. They performed well in user tests for languages including Japanese, Brazilian Portuguese, Vietnamese, Modern Standard Arabic, Mexican Spanish, and Hindi.
The models are currently available via Google AI Studio and the Gemini API. Several platforms, including Agora, LiveKit, Pipecat, and Vercel, have already announced integration with the new technology. The Flash-Lite version is specifically optimized for large-scale projects such as dubbing and the development of voice-based AI agents.
