Gemini 3.8 Flash TTS

Gemini 3.8 Flash TTS is Google's text-to-speech model for expressive, natural-sounding speech from text. It follows natural-language delivery directions for tone, pace, and emotion, renders inline tags such as laughs, sighs, and pauses, and generates two-speaker dialogue in a single request. It detects the language of the text automatically across more than 130 languages, making it suitable for narration, voice agents, podcasts, audiobooks, and localized voiceover.

Complete technical specification for integration
Ready-to-use code snippets for common workflows
Step-by-step tutorials for advanced use cases
← All GuidesInline tags
How to add laughs, sighs, breaths and pauses to Gemini 3.8 Flash TTS speech with inline tags: the syntax, the documented tags and where to place them.
Introduction
Inline tags add a sound or a pause at an exact point in the speech. You write them in angle brackets inside speech.text, such as <laugh> or <short pause>, and the model performs them instead of reading them: it makes the sound or leaves the silence where the tag sits, then carries on with the text.
<yawn> Morning, everyone. Okay... coffee first. <chuckle> Then I'm finally showing you the studio we spent all summer building.
This guide covers the request, the documented tags, pauses, breath and effort, laughs and reactions, tags with a style and tags in other languages.
Gemini 3.1 Flash TTS writes these tags in square brackets. Gemini 3.8 Flash TTS uses angle brackets, and square brackets mark the speakers of a dialogue.
The request
A tag goes exactly where the sound happens, between the words around it, and the text after it continues in the same delivery.
import { createClient } from '@runware/sdk'
const client = await createClient({ apiKey: process.env.RUNWARE_API_KEY })
await client.connect()
const [result] = await client.run({
model: 'google:gemini@3.8-flash-tts',
speech: {
text: '<yawn> Morning, everyone. Okay... coffee first. <chuckle> Then I\'m finally showing you the studio we spent all summer building.',
voice: 'Callirrhoe'
}
})import asyncio
import os
from runware import Runware
async def main():
async with Runware(api_key=os.environ["RUNWARE_API_KEY"]) as client:
results = await client.run({
"model": "google:gemini@3.8-flash-tts",
"speech": {
"text": "<yawn> Morning, everyone. Okay... coffee first. <chuckle> Then I'm finally showing you the studio we spent all summer building.",
"voice": "Callirrhoe"
}
})
asyncio.run(main())curl https://api.runware.ai/v1 \
-H "Authorization: Bearer $RUNWARE_API_KEY" \
-H "Content-Type: application/json" \
-d '[
{
"taskType": "audioInference",
"taskUUID": "2b1084c3-4780-4106-882a-948456fda422",
"model": "google:gemini@3.8-flash-tts",
"speech": {
"text": "<yawn> Morning, everyone. Okay... coffee first. <chuckle> Then I'm finally showing you the studio we spent all summer building.",
"voice": "Callirrhoe"
}
}
]'runware run google:gemini@3.8-flash-tts \
speech.text="<yawn> Morning, everyone. Okay... coffee first. <chuckle> Then I'm finally showing you the studio we spent all summer building." \
speech.voice=Callirrhoe{
"taskType": "audioInference",
"taskUUID": "2b1084c3-4780-4106-882a-948456fda422",
"model": "google:gemini@3.8-flash-tts",
"speech": {
"text": "<yawn> Morning, everyone. Okay... coffee first. <chuckle> Then I'm finally showing you the studio we spent all summer building.",
"voice": "Callirrhoe"
}
}Response
[
{
"taskType": "audioInference",
"taskUUID": "2b1084c3-4780-4106-882a-948456fda422",
"audioUUID": "2ad5b30a-aadb-4d88-9df6-75b9073c0063",
"audioURL": "https://am.runware.ai/audio/os/a14d18/ws/2/ai/2ad5b30a-aadb-4d88-9df6-75b9073c0063.mp3"
}
]Pauses
<short pause> and <long pause> leave a silence where the listener expects a beat, such as before a reveal or after a question. Punctuation paces the read as well, as Prompting covers, and a tag makes the pause part of the performance.
And this week's top scorer, with four thousand two hundred points, is... <long pause> Priya!
A voice agent can use a short pause between two pieces of information, so the customer hears the total before the question.
Your total comes to forty-two fifty. <short pause> Would you like the receipt sent to your email?
Breath and effort
Breath tags make physical effort audible, which suits workout coaching and sports content, where a perfectly smooth read sounds like it was recorded sitting down.
Last round. <heavy breath> Ten more squats, chest up, push through your heels. <exhales> Good. Five, four, three, two... and done.
Laughs and reactions
A reaction tag puts a response into the speaker's voice, such as surprise at an unboxing or a groan at bad news.
<gasp> Okay, it's even smaller than I thought. <laugh> It actually fits in my jacket pocket.
<sigh> So I finally checked my screen time. Six hours. <groan> On a Tuesday.
Place a reaction where a person would make it. A sigh opens a resigned line, and a laugh follows the line it reacts to.
Tips
-
Put the tag where a person would make the sound. Before the line for a sigh, after the joke for a laugh.
-
Use the documented tags. They are the ones Google describes for the model, and they all stand for a sound a person makes.
-
Keep tags in English. The text can be in any language the model reads.
-
Use a style for a mood. A tag is one moment, and a delivery that lasts belongs in
settings.style.