Gemini 3.8 text‑to‑speech says hello

Google has unveiled the Gemini 3.8 Flash and Flash‑Lite text‑to‑speech (TTS) models, promising creators a new level of expressive, customizable voice generation that can be built from scratch and scaled for high‑volume audio production.

What’s new in Gemini 3.8 Flash TTS

The Flash model delivers ultra‑low latency synthesis while preserving naturalness, making it suitable for real‑time applications such as interactive assistants, live streaming, and gaming. Flash‑Lite, a lighter variant, offers comparable quality with an even smaller footprint, enabling deployment on edge devices and mobile platforms.

Custom voice creation

Both models support end‑to‑end voice customization, allowing developers to train a unique voice using a modest dataset of recorded speech. The system then extrapolates the speaker’s style, tone, and prosody, producing consistent output across diverse scripts.

Google’s approach leverages a large multilingual foundation model, so the custom voices inherit multilingual capabilities and can seamlessly switch languages without retraining.

Scalable high‑volume generation

Flash’s architecture is optimized for batch processing, enabling enterprises to generate thousands of audio clips per minute while maintaining low latency. This opens new possibilities for large‑scale content creation, such as automated podcast production, audiobooks, and personalized marketing messages.

• Real‑time voice assistants
• Live‑stream commentary
• Mass‑customized audiobooks
• Dynamic advertising spots

“Gemini 3.8 Flash gives us the speed we need for interactive experiences without sacrificing the natural feel of human speech,” said a Google AI spokesperson.

Developers can access the models through Google Cloud’s Vertex AI platform, where they can fine‑tune voices, manage usage quotas, and integrate the service via simple API calls.

For more details, see Google AI blog’s coverage of Gemini 3.8 text‑to‑speech.