Gemini 3.8 TTS: speech_metadata for a whole turn, tags for one moment

Gemini 3.8 Flash TTS has two style controls: a speech_metadata annotation that sets delivery for a whole turn, and inline angle-bracket tags such as cough or pause for a single moment. Use metadata for the character's overall manner and tags for one-off sounds; Sume's TTS job takes plain transcript text and documents no equivalent tag syntax.

What Google's speech docs say

The Gemini API speech page describes two models, gemini-3.8-flash-tts and gemini-3.8-flash-lite-tts, and these controls:

• Turn-level styling through a speech_metadata annotation, for sustained delivery.
• Inline angle-bracket tags for momentary vocal events, for example a cough or a pause.
• Single-request multi-speaker generation supports up to 2 speakers with prebuilt voices.
• 30 prebuilt studio voices, plus an extended library.

Which control for which job

Match the control to the length of the effect.

What the Sume job takes

Sume's tts_create takes a transcript, a voice selector, a language, an output format, word timestamps and sentence segmentation. Completed jobs record the synthesis settings (generation_config and speed), each null when you did not send it. The tool description lists no tag syntax, so write delivery into the text and the voice choice.

Sume's rate card is $0.0475 per 1,000 characters. Gemini prices audio output per million tokens: $9.00 through December 31, 2026, then $18.00 for Flash TTS, per Google's pricing page.

A workable rule

If you need acted lines with sighs and laughs, test Gemini's tags on your script. If you need a clean, repeatable narration with word timings and slices for a timeline, a Sume job gives you the file and the timings in one result.