Gemini 3.8 TTS Playground

Google has unveiled two fresh text-to-speech models for its Gemini AI - gemini-3.8-flash-tts and gemini-3.8-flash-lite-tts. These models are accompanied by a vast library of more than 2,000 voices. Additionally, users can create customized voices using just a 30-second audio sample of their own voice or a voice they have rights to use.

I developed a playground interface for these models called bring-your-own-key, leveraging the GPT-6 Astra model and taking advantage of the Gemini API's open CORS policy. The API facilitates the creation of full conversations between multiple characters, each with distinct voices and voice style instructions. I've included a brief demo clip of a conversation between two pelicans discussing whether they should relocate to the Pacifica Pier.

I generated the script using Claude 4.5 Opus and rendered it using the tool. The audio generation took around 20 seconds using Gemini 3.8 Flash TTS, costing $2.74. This demonstration was created by Simon Willison on September 23rd, 2026. For $10 a month, I'll provide you with a curated email newsletter featuring the month's most significant developments in large language models.