Google offers its Flash range in text-to-speech, with dialogue with several speakers, fine regional variations and laughter or sigh tags inserted into the text.
On September 24, Google released two models dedicated to text-to-speech, Gemini 3.8 Flash TTS and its lightweight Flash-Lite TTS version. The offer is distinguished less by the raw quality than by the breadth of the catalog: more than 2,000 pre-built voice profiles covering fine regional variations, on more than a hundred languages — Quebec French, Scottish English, Mexican Spanish, Brazilian Portuguese, Japanese, Vietnamese, Modern Standard Arabic or Hindi.
The models handle multi-speaker dialogue, generating long audio sequences, and inserting nonverbal acoustic markers directly into the text, via tags such as <laughs> or <sigh>. The target is explicitly high-volume audio production, where the script must drive the interpretation without going back to a studio.
On the Hume AI Voice Design Benchmark, Gemini 3.8 Flash TTS scores 71.4 overall and 60.8 on accent modeling, which scores best in its class on this criterion. Access is through Google AI Studio and the Gemini API, with integrations at Figma, HeyGen, and Wondercraft; Gemini Enterprise deployments will receive admin access to the API in a later wave.
Read the full article on the original source
Go to the source →