Gemini 3.1 Flash TTS: Next-Generation Expressive AI Voice

Gemini 3.1 Flash TTS has been released, offering greater controllability, expressiveness, and audio quality. It's designed to let developers, enterprises, and everyday users build more sophisticated AI voice applications, and is currently rolling out in preview across Gemini API, Google AI Studio, Vertex AI, and Google Vids.

Voice quality has also been significantly improved. Google states that this is its most natural and expressive model to date, achieving an Elo of 1,211 on the Artificial Analysis TTS leaderboard. The company also highlighted that it supports native multi-speaker dialogue, works across 70+ languages, and sits at a strong point in the quality-to-cost balance.

The biggest change is audio tags. Natural-language instructions can be embedded directly within text input to finely control speaking style, pace, and delivery, and Google AI Studio provides a more intuitive developer experience for handling them.

Key features include the following.

• Scene direction: Insert scene and dialogue instructions to maintain a character's context and reactions
• Speaker-level specificity: Fine-tune tone, pace, and accent using Audio Profiles and Director's Notes
• Seamless export: Export tuned parameters as Gemini API code to reuse consistent voices

Enterprise use cases are also being targeted. Google explained that these control features are especially useful for products, characters, interactive experiences, and services where localization matters, and noted that users can experiment directly in the Google AI Studio Playground.

Alongside the rollout, SynthID watermarking is also applied. All audio generated by Gemini 3.1 Flash TTS includes an invisible watermark that allows AI-generated content to be identified, presented as a safeguard to help reduce the spread of misinformation.