Laugh and sigh tags in TTS: Gemini 3.8 tags vs Sume's emotion field

Gemini 3.8 TTS documents inline vocal tags in the transcript, and its launch post describes scripted vocal bursts and backchanneling through non-verbal cues. Sume TTS documents three delivery controls instead: speed, volume and a short emotion string. Nothing in the Sume OpenAPI promises inline tags, so test one short job before relying on them.

What does each side document?

The Gemini row is from Google's speech generation page, read 2026-10-09. The Sume row is from the OpenAPI generation_config schema.

How do I test a tag on Sume without wasting a script?

Run a 40-character test with the tag in place and listen. A tag the engine does not understand may be read aloud as words, or dropped. That costs 1 cent, and the character count includes the tag, because the API bills spaces and punctuation as characters.

If the tag does not work, move the intent to the emotion field and the punctuation. A line such as "Oh. Well, that is that." with a short emotion guide carries a sigh without any markup.

What about stressing one word?

Gemini's own page shows capital letters as the way to put natural vocal stress on key words. That is a Gemini behaviour; the Sume docs do not describe it, so check it the same way, with one short job.

For a direct comparison of Gemini's turn style and Eleven's tags, see the style field comparison.