Google Launches Gemini 3.8 Flash TTS and Flash-Lite TTS Models

Google launched two text-to-speech models on September 23, Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS, and its announcement splits them by job. Flash TTS is for production work where a person directs each line, such as games, audiobooks and podcasts, and for designing custom voices from scratch. Flash-Lite TTS adjusts its own tone and pacing on the fly and is aimed at near real-time voice agents and high-volume work such as dubbing and bulk audio, where Google stresses cost-efficient scale.

Both models are directed through the script itself. Google says users can guide delivery line by line, stage back-and-forth conversations, and write cues into the text, such as a laugh tag or an active-listening interjection like mhm, while generating hours of consistent audio. Voices come in more than 100 languages, either custom-built or chosen from more than 2,000 ready-made ones. Writing cues into the script is not new to this generation: Google introduced audio tags with Gemini 3.1 Flash TTS in April. The launch video also shows a user recreating their own voice, which the narration says is protected by a quick verbal identity check; only an automatic transcript of that video was available, so how the check works is not covered here.

The announcement gives no prices, benchmark results or availability details. What it supports is a choice between control and speed: Flash TTS when someone wants to direct a performance line by line, and Flash-Lite TTS when a system has to speak quickly and cheaply at volume.