Google has introduced Gemini 3.1 Flash TTS, a new text-to-speech model designed to make synthetic voices sound more natural, expressive, and easier to control.
Gemini 3.1 Flash TTS is already making an impact on industry benchmarks. According to Artificial Analysis, the model scored 1,211 Elo points on its TTS leaderboard, which is based on blind listening tests involving thousands of human comparisons.
That places it second globally, just behind Inworld TTS 1.5 Max (1,215), and ahead of ElevenLabs Eleven v3 (1,179). Artificial Analysis also categorized the model in its “most attractive quadrant,” highlighting its balance between high-quality speech output and relatively low cost.
One of the standout features in Gemini 3.1 Flash TTS is the introduction of audio tags, a system that lets users control how speech is delivered with simple text instructions.
Developers can insert tags directly into scripts to adjust tone, pacing, and expression in real time.
According to reports, the model supports over 200 such tags, offering a level of control not commonly seen in traditional TTS systems. This inline prompting approach allows users to shape speech output without complex audio engineering, making it easier to experiment with and refine voice experiences.
The model supports more than 70 languages, along with regional accent variations. This includes multiple English accents, ranging from American styles to British variants like RP and Brixton, as well as a wide mix of global languages. For Workspace users, integration with Google Vids adds 30 conversational voice options across 24 languages, expanding accessibility and localization for content creators and businesses.
Safety measures with built-in watermarking
To address concerns around AI-generated media, all audio produced by Gemini 3.1 Flash TTS is embedded with SynthID watermarking. According to the company’s announcement, “This imperceptible watermark is interwoven directly into the audio output, allowing the reliable detection of AI-generated content to help prevent misinformation.”
This watermark is not audible but can be detected to verify whether a clip was generated by AI. Google says this approach is intended to help identify synthetic content and reduce the risk of misinformation.
How to access it
Gemini 3.1 Flash TTS is available now. Developers can access it in preview through the Gemini API and Google AI Studio. Enterprise teams can test it on Vertex AI. And Workspace users will find it inside Google Vids.
Also read: Google’s March AI push expanded Gemini with more proactive tools, deeper personalization, and new options for creators and developers.