Gemini 3.8 Text-to-Speech Sandbox

If you've been waiting for a flexible, affordable way to turn text into natural-sounding speech, Google just gave you two new reasons to pay attention. Today the company released two new Gemini text-to-speech models: gemini-3.8-flash-tts and gemini-3.8-flash-lite-tts. In this article, we'll walk through what these models offer, how a developer built a quick playground to test them, and what the real-world speed and cost look like based on a hands-on demo.

A Massive Voice Library and Custom Cloning

The headline feature of both models is the sheer range of voices available. Google ships them with a library of more than 2,000 voices, which means you can almost certainly find a tone, accent, or speaking style that fits your project without any extra work.

If the built-in voices aren't quite right, there's another option: custom voice creation. According to Google, all it takes is a 30-second audio sample of your own voice — or a voice you have the rights to use. That opens the door for branded audio, personalized assistants, or consistent narration across long pieces of content, as long as you stay on the right side of the consent and licensing rules.

Multi-Speaker Conversations Made Simple

One of the more interesting capabilities of the Gemini TTS API is how easily it handles dialogue. Rather than generating each speaker's lines separately and stitching them together, the API lets you define an entire conversation between multiple characters in one go — with each character assigned a different voice and its own voice style instructions.

That means you can specify not just who is talking, but how they're talking: excited, deadpan, hesitant, cheerful, and so on. For anyone building podcasts, audio dramas, or interactive applications with distinct characters, this is a significant time-saver compared to juggling multiple single-speaker requests.

A Vibe-Coded Playground for Testing

To explore what these models can do, one developer built a bring-your-own-key playground interface using what they describe as "vibe coding" with GPT-6 Astra. The interface takes advantage of the underlying Gemini API's open CORS policy, allowing the tool to call the API directly from the browser without needing a backend server.

The result is a lightweight way for anyone with their own API key to experiment with the models, test different voices, and hear results immediately — no infrastructure required.

Putting It to the Test: Two Pelicans Debate a Pier

To demonstrate the multi-speaker feature, the developer created a short audio clip of a conversation between two pelicans debating whether they should move to the Pacifica Pier. The script for the exchange was written by Claude 4.5 Opus, which also generated a URL to render the dialogue using the playground tool.

The numbers from this demo are worth noting. The clip runs 1 minute and 18 seconds long, and it took roughly 20 seconds to generate using the standard Gemini 3.8 Flash TTS model — not the cheaper Flash-Lite variant. The total cost came to 2.74 cents.

What This Means for Developers and Creators

For anyone comparing options in the text-to-speech space, a few things stand out from this release:

• Choice between tiers: The Flash-Lite model gives you a lower-cost option when quality demands are modest, while the standard Flash model is there for more demanding output — and in this test, even the standard tier produced over a minute of audio for under three cents.
• Speed: Turning around nearly a minute and a half of audio in about 20 seconds makes iterative workflows practical. You can tweak a script, regenerate, and listen again without long waits.
• Dialogue as a first-class feature: Defining multi-character conversations with distinct voices directly in the API removes a layer of assembly work that most TTS workflows still require.

As always with newly released models, it's worth running your own tests on your specific content before committing to one tier or the other. Voice quality can vary depending on the language, tone, and complexity of the material you feed it. But based on this early hands-on look, Google's latest Gemini TTS models look like a strong, budget-friendly entry point for voice generation projects of all sizes — from quick prototypes to full conversational audio.