Google DeepMind releases Gemini 3.8 Flash TTS and Flash-Lite TTS for expressive audio generation

Google DeepMind has introduced two new text-to-speech models, Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS, as part of its Gemini Audio family. These models are positioned as the most expressive audio generation tools yet, enabling users to create custom character voices and direct scene dialogue with granular control over performance elements such as pacing, emotion, and conversational nuances.

The Gemini 3.8 Flash TTS model is designed for deep creative direction, allowing users to generate entirely new voices from scratch using natural language prompts. It supports customization across more than 100 languages and dialects, enabling creators to craft voices tailored to specific roles, accents, and vocal characteristics. For example, the model can produce a high-energy DJ voice from Melbourne, a super-tinny robot voice, or a Japanese dragon with lifelike qualities.

Gemini 3.8 Flash-Lite TTS is optimized for high-volume, cost-efficient applications such as dubbing, audio content creation, and expressive voice agents. It provides fine-grained control over tone, pacing, and expressive nuance, making it suitable for scalable deployments where performance and cost efficiency are priorities.

The models complement Google DeepMind’s existing Gemini Audio family, which includes 3.5 Live Translate, 3.5 Transcribe, 3.8 Live, and 3.8 Live Extended Thinking. Both new models are accessible through Google AI Studio, the Gemini API, Gemini Enterprise, Gemini Notebook, and Google Vids, broadening their availability to developers, creators, and enterprises.

Gemini 3.8 Flash TTS includes an expansive voice library featuring 2,000+ production-ready voices with broad language coverage, including regional varieties such as Mexican Spanish, Quebec French, and Scots English. The model also supports voice replication, allowing users to recreate consistent vocal profiles from a 30-second audio sample of a target voice or a voice they have the rights to use. This feature is backed by built-in consent verification, SynthID watermarking, and C2PA credentials to protect both developers and vocal talent.

Users can save and manage custom voices to ensure consistent performance and minimal drift across ongoing projects. A voice remixing feature is also planned, enabling users to fine-tune timbre, pitch, pace, and accent of existing voices using prompts such as “add subtle Southern US accent” or “soften the delivery.”

The models provide line-by-line control over audio delivery, allowing users to direct performances with precision. This includes writing stage directions or leveraging natural script cues to achieve desired emotional tones, from calm customer service interactions to whispered suspense scenes.

Built-in safety tools such as watermarking are included to help secure generated audio and mitigate risks associated with synthetic media. The announcement emphasizes responsible use through consent verification and technical safeguards embedded in the models.