Intel / article

Gemini 3.8 TTS Playground

Source: simonwillison.net
Date: Published Sep 23, 2026

Context

Simon Willison's speech playground lets users try Google's Gemini 3.8 text-to-speech models with their own API key. Users can compose narration or a conversation, give speakers different voices and delivery instructions, then preview the generated audio. Composition settings can be saved in shareable URLs.

That makes it a way to experiment with how a script sounds before integrating speech generation into an application. It complements Vercel's AI Gateway integration, which exposes the same model family to application code through one API.

Key takeaways · AI-distilled

Simon Willison built the playground as a bring-your-own-key browser app, which works because the Gemini API has an open CORS policy, letting a page call it directly with the user's key. Compose settings save to bookmarkable URLs for sharing.
Willison highlights that a single API request can define a full conversation between multiple characters, each with a different voice and its own voice style instructions.
In Willison's test, Gemini 3.8 Flash TTS (not the cheaper Flash-Lite) took about 20 seconds to generate 1 minute 18 seconds of two-speaker audio, at a cost of 2.74 cents.

Read the source simonwillison.net