In a move reflecting the rapid advancement in the world of generative artificial intelligence, Google announced on Wednesday the launch of the “Gemini 3.1 Flash” text-to-speech model, which the company describes as its most expressive and controllable model to date.
This launch underscores the company’s commitment to providing advanced tools for developers and content creators. The new model is now available in beta via the Gemini API, Google AI Studio, Vertex AI, and the Google Vids app for Google Workspace users.
Precise and advanced control via audio tags
One of the standout features of the new model is the ability to precisely control audio outputs, offering more than 200 audio tags that developers can integrate directly into text inputs. These tags allow steering of vocal style, speaking rate, accent, and emotional expression with an unprecedented level of accuracy.
These tags range from complex emotions like “determination” and “curiosity” to directing delivery style by adding natural effects such as “whispers” and “laughter.” Google calls this capability the “compositional approach” to voice generation, enabling users to shape a vocal persona like a theater director guiding actors, opening wide horizons for creating rich, lifelike audio content that goes beyond mere robotic text reading.
Multilingualism and complex conversation support
In terms of linguistic diversity, Gemini 3.1 Flash is designed to be a truly global tool, supporting over 70 different languages, including widely spoken ones like Hindi, Japanese, and German, serving a massive global user base. To make getting started easier, the model provides 30 pre-made voices that developers can use as starting points to develop their custom voices.
Furthermore, the model features native, built-in capability to handle dialogues involving multiple speakers. This revolutionary feature maintains the natural flow of conversations without requiring separate API calls for each individual voice. This capability specifically targets podcast creators, dramatic scriptwriters, and smart voice assistant interface developers, helping significantly reduce software complexity and accelerate production.
Performance superiority and top global rankings
The new model’s achievements are not limited to technical features; it has also proven its worth in independent performance benchmarks. According to Google AI Studio, the model achieved a score of 1,211 points on the Artificial Analysis text-to-speech leaderboard.
The Artificial Analysis platform noted that Gemini 3.1 Flash secured the second spot on its audio arena leaderboard, outperforming ElevenLabs’ well-known version 3 model, clear evidence of the quality of Google’s audio output and its strong ability to compete in this fast-growing market.
Content reliability through watermarking technology
Amid growing concerns regarding the misuse of AI technologies and synthetic media generation, Google has placed great emphasis on safety and reliability. A digital watermark is embedded in all audio clips generated by the model using SynthID technology. This innovative Google technology acts as an invisible or inaudible human fingerprint, specifically designed to identify AI-generated content to help prevent the spread of misinformation.
The company emphasizes that embedding this watermark is done professionally and advancedly so that it does not degrade audio quality whatsoever, ensuring users receive the highest possible audio fidelity while simultaneously maintaining safety and transparency standards.
Technical access details and usage limits
For developers and companies wanting to explore the new model’s capabilities, it can be accessed via the dedicated beta program identifier through the Gemini API. The model comes with a maximum input token limit of 8,192 tokens, while the output token limit reaches 16,384 tokens, providing ample space to handle long and complex texts in a single session.
This important launch follows closely after Google’s rollout of another model on March 25—Gemini 3.1 Live Flash—built specifically for live-audio AI applications requiring real-time dialogue processing. This completes Google’s audio ecosystem, delivering comprehensive solutions that meet all the needs of the contemporary tech market.
Frequently Asked Questions
What is the new model launched by Google for text-to-speech?
Google launched Gemini 3.1 Flash, an advanced text-to-speech system featuring exceptional expressive capability and allowing developers to control vocal performance through more than 200 specialized audio tags.
How many languages does the model support?
The model supports more than 70 global languages, such as Hindi, Japanese, and German, and includes 30 pre-made voices ready for immediate use by content creators.
How does Google handle misinformation concerns with this model?
Google has integrated SynthID technology into the model, an imperceptible watermark embedded into the audio clip to recognize AI-generated content without affecting audio quality or clarity.
What usage limits are available to developers in the model?
The model provides a large data capacity, with a maximum input limit of 8,192 tokens, while the maximum output limit reaches 16,384 tokens, enhancing its efficiency in processing long dialogues.
