Expressive tags and speech control

Updated November 3, 2026, based on the xAI Text to Speech page and Voice API reference. The tags below follow the official list: 14 inline tags and 12 wrapping tags. Sources: Text to Speech · Voice API.

Beyond flat delivery

Left to itself, a speech engine reads everything in the same tone, including the punchline of a story or the key warning in the middle of a user guide. Grok TTS lets you take back control with speech tags, which shape pace, volume, pitch, and style, and can even insert laughter, sighs, or breaths. There are two distinct mechanisms, and mixing them up is the number one cause of disappointing results: inline tags trigger a one-off effect, while wrapping tags change how a whole passage is read.

Inline tags: one effect, one spot

You write them in square brackets, and they act exactly where you place them without changing the text that follows.

Pauses are by far the most useful day to day, because they add the spoken punctuation that written text lacks. Compare an instruction rattled off in one go with the same instruction given room to breathe:

Welcome to our platform. [pause] Today, we'll explore the main features. [long-pause] Let's start with signing up.

Vocal expressions, on the other hand, call for a light touch. A well-placed [laugh] makes a sentence feel friendly; three in the same paragraph make you sound like a nervous host.

That's a great question! [laugh] The answer might surprise you. [pause]

Wrapping tags: one style, one stretch of text

These tags surround the text they modify, as in <tag>text</tag>, and their effect lasts exactly as long as the wrapped passage.

Contrasting volume lets you rank information without adding a word: what’s said softly sounds like a confidence, what’s said loudly like an alert.

Listen carefully: <soft>this is a secret</soft>. But <loud>this is important!</loud>

Speed does the same for attention. Speeding up suggests urgency or an aside; slowing down signals a sentence worth remembering. That’s a valuable habit in a training module, where you always slow down on the definition of a new term.

Now I'm going to <fast>talk really fast to show urgency</fast> and then <slow>slow down to stress this point</slow>.

Whispering creates instant complicity, as long as it stays the exception in the paragraph.

<whisper>Don't tell anyone</whisper>, but the new version comes out tomorrow!

Combining both

Nothing stops you from mixing the two families in the same text, and that’s where synthesis starts to sound like a performance. The excerpt below strings together a breath, pauses, a softer passage, slowing down, a whisper, a laugh, and a rise in volume:

[breath] Hello, everyone. [pause]
Today, I'm going to tell you <soft>an extraordinary story</soft>. [long-pause]
Once upon a time, <slow>in a faraway land</slow>, a developer discovered [pause] <whisper>the secret of perfect speech synthesis</whisper>. [laugh]
<loud>And everything changed!</loud>

Read this example aloud, following the cues, and you’ll hear that the tag density is already at the edge of reasonable. Across a whole audiobook, that much would be exhausting. The rule of thumb is to tag the moments that carry intent and let the rest breathe.

Where these tags really make a difference

In audiobook narration, pauses, whispers, and changes of pace are what keep listeners awake through an hour of listening. For a voice assistant, a simple [pause] between two pieces of information gives users time to take in one before the next. Educational content benefits from combining <emphasis> on key terms with <slow> on difficult passages, exactly as a trainer would in front of a class. Audio marketing uses <build-intensity> to build momentum and <loud> on the call to action. And for video game characters, combining <whisper>, <lower-pitch>, and [laugh] gives identical lines of dialogue very different personalities.

One last practical point: fine-tune your tagging with individual HTTP requests, which are easy to replay and correct, then check how the approved text sounds in your streaming pipeline. xAI also recommends wrapping complete sentences rather than isolated words, and pairing tags with punctuation instead of stacking them.

Key takeaways

• Inline [tag] tags insert one-off effects (pauses, laughter, sighs)
• Wrapping <tag>text</tag> tags change the style of a passage (volume, speed, tone)
• You can combine tags to create rich, expressive narration
• xAI publishes 14 inline tags and 12 wrapping tags: stick to that list