Audio
Gemini 3.8 Flash TTS in Twelver: two-voice dialogue and delivery directed in plain words
We’ve moved Google speech to Gemini 3.8 Flash TTS. The tools are the same — a single voice, a two-voice dialogue, a voice from a description. Delivery direction now travels separately from the text, so the model follows it but never reads it aloud.
What’s new
- Two-speaker dialogue — different voices, one file, one call.
- Describe the delivery in plain words: “slow, warm, almost a whisper by the end”.
- Laughs, sighs and pauses go in as <laugh>, <sigh>, <short pause> tags right in the text — they come out as sound, not as words.
- Dozens of languages — Turkish, Arabic and Hindi included, which Qwen does not speak.
- A second of finished speech costs about 10% less than on the previous version.
At a glance
| Detail | Value |
|---|---|
| Model | Gemini 3.8 Flash TTS (Google) |
| Replaces | Gemini 3.1 Flash TTS (preview) |
| Modes | Single voice · two-voice dialogue · voice from a description |
| Languages | Dozens |
| Text per request | Up to 7,000 characters |
| Billing | Per second of finished audio |
| Where in Twelver | Chat · MCP server |
Two voices in one dialogue
One request, and two podcast hosts talk to each other with natural pauses between turns. Not two files stitched together in an edit.
A language Qwen does not speak
When Gemini is worth it and when the cheaper Qwen3-TTS does the job is in the Gemini TTS chapter.
Frequently asked questions
What’s new in Gemini 3.8 Flash TTS?
It is the stable release in place of the preview. Delivery direction now travels separately from the text, so the model follows it without reading it aloud, and a second of speech costs about 10% less. The capabilities are the same: a single voice, a two-voice dialogue and a voice from a description.
How do I voice a dialogue between two characters?
Write the lines of two speakers by role, and the model records them in different voices in one file. You can try it in the Gemini TTS chapter, which has a window with the model itself.
Gemini TTS or Qwen3-TTS?
For ordinary narration Qwen3-TTS is roughly four times cheaper and can clone your voice. Take Gemini for two-voice dialogue and for languages outside the ten Qwen speaks.
Can I call Gemini 3.8 Flash TTS from Claude or Cursor?
Yes, over MCP: text-to-speech is available through MCP in Claude, Cursor and other clients — see the MCP page.
Try it
Open in Twelver chatRate your experience