Google has introduced two new text-to-speech models, Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS, which the company says are built for expressive, high-quality speech generation at global scale. The company says the pair is designed for creators, developers, and enterprises building multilingual voice experiences.
The headline claim is benchmark performance. Google says Gemini 3.8 Flash TTS took the number one overall spot on Hume AI's Voice Design Benchmark with a score of 71.4, and also led in accent modeling at 60.8. On Hume AI's Overall Quality Index, the company says the two models ranked first and second, respectively. Those are Google's characterizations of third-party benchmark results, not independently verified findings.
Read more: Google Debuts Gemini 3.8 Live Voice Models for Developers
Google says the new models improve on Gemini 3.1 Flash TTS across a range of use cases, including long-form content and dual-speaker screenplay control. The company also points to blind human preference evaluations on Voice Arena, where it says both models placed near the top among competitors in several key global languages. Those languages include Japanese, Brazilian Portuguese, Vietnamese, Modern Standard Arabic, Mexican Spanish, and Hindi.
Language coverage is the other main selling point. Google says the models support more than 100 languages. That breadth is aimed at developers and enterprises that need to produce voice content for audiences across multiple regions rather than a single market.
The announcement arrives as Google continues to fold Gemini branding into its hardware and software lineup. A Googlebook described as designed for Gemini Intelligence was announced two days earlier. The company has not disclosed pricing for either text-to-speech model, and the notice does not say whether the models are generally available or in preview.
For anyone weighing the benchmark claims, the numbers come from Hume AI's evaluations as cited by Google, and the Voice Arena results are described as blind human preference tests. Google frames the models as delivering expressive performances without giving up reliability.
