# Emotion & tone
URL: /tts/concepts/emotion-and-tone
Shape Soniox Text-to-Speech delivery with audio tags, text formatting, and punctuation to make speech expressive and emotional.
## Overview
Soniox Text-to-Speech lets you control *how* the words are delivered, not just which words are spoken. You have two toolkits, and they combine freely:
* **[Audio tags](#audio-tags)** — bracketed cues like `[excited]`, `[whispering]`, or `[laughs]` that control emotion, tone, human sounds, and delivery.
* **[Text formatting & punctuation](#text-formatting--punctuation)** — capitalization, emphasis, elongation, and punctuation marks written into the words themselves (`UPPERCASE`, `*stress*`, `sooo`, `...`, `—`, `?!`).
Both are written directly into the `text` you send via API:
```json
{
"model": "tts-rt-v2",
"language": "en",
"voice": "Adrian",
"audio_format": "wav",
"text": "[excited] INCREDIBLE news! [calm] Let me explain what happened."
}
```
These controls shape the *expressive* layer: emotion, sounds, emphasis, and
local pacing. To change the overall speaking rate use the [`speed` parameter](/tts/concepts/speech-speed)
instead.
## Audio tags
Place a bracketed tag before the text you want to affect to set *how* it is delivered. The tag describes the feeling or manner; the words after it inherit that delivery.
The tags listed below are not exhaustive. Other tags may also work — test them before relying on them in production.
Always write audio tags in **English**, regardless of the text language.
Localized tags (for example, `[emocionado]` instead of `[excited]`) are not
supported.
```text
[excited] ¡Tengo noticias INCREÍBLES! [calm] Déjame explicarte lo que pasó.
```
### Emotion & tone
| Category | Tags |
| ------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------- |
| Emotion | `[happy]` `[sad]` `[angry]` `[excited]` `[nervous]` `[fearful]` `[surprised]` `[annoyed]` `[relieved]` `[disappointed]` `[curious]` `[delighted]` `[calm]` |
| Tone & manner | `[warm]` `[stern]` `[serious]` `[playful]` `[sarcastic]` `[flirty]` `[deadpan]` `[sincerely]` `[reassuringly]` `[dramatically]` `[mockingly]` |
```text
[excited] INCREDIBLE news! [calm] Let me explain what happened.
[sad] I don't know what to say anymore.
[warm] [softly] Hey. [reassuringly] It's okay.
```
### Human sounds
Non-lexical actions you can't spell as a word go in brackets:
| Category | Tags |
| --------------------- | -------------------------------------- |
| Amusement | `[laughs]` `[chuckles]` `[giggles]` |
| Sadness | `[sobs]` `[sniffles]` `[whimpers]` |
| Relief / exasperation | `[sighs]` `[exhales]` |
| Surprise | `[gasps]` |
| Effort / pain | `[grunts]` `[groans]` |
| Physiological | `[clears throat]` `[coughs]` `[yawns]` |
```text
[chuckles] And I just let go — [laughs] best thing all week.
[sighs] I really thought I'd get the job. [sniffles] Oh well.
```
Spoken interjections like `Hmm`, `Um`, or `Mhm` are different: write them as ordinary words, not tags — see [Spoken interjections](#spoken-interjections).
Sound effects related to the environment (`[dog barks]`, `[phone rings]`) are
not supported.
### Volume, pace & pitch
Tags that control the mechanics of delivery: how loud, how fast, and at what pitch.
| Category | Tags |
| ------------- | ----------------------------------------------------------------------------------------------------------------------- |
| Volume | `[whispering]` `[softly]` `[loudly]` `[shouting]` `[getting louder]` `[getting quieter]` `[trailing off]` `[muttering]` |
| Pace & rhythm | `[slowly]` `[quickly]` `[rushed]` `[drawn out]` `[hesitantly]` |
| Pitch | `[high-pitched]` `[low voice]` `[monotone]` |
| Voice quality | `[breathy]` `[raspy]` `[trembling voice]` `[shaky]` |
```text
[whispering] Something's out there. [slowly] Do not move.
[shouting] Get out of there!
[low voice] Sit. [hesitantly] We need to talk.
```
### Pauses
Use tags to insert deliberate silence:
* `[pause]` — the default pause; use it for most cases.
* `[long pause]` — a longer, more dramatic gap; use occasionally.
For shorter beats inside a sentence, use [punctuation](#punctuation) (`...`, `,`) instead.
```text
So, here's the plan. [pause] Let me walk you through it.
```
### Tag limitations
Tags work best when they stay close to the short, single-purpose forms listed above.
* **Keep each tag simple.** `[laughs]` and `[shouting]` are reliable; compound directions like `[laughs and screams really loud]` or `[says the next line sarcastically]` are not. Rewrite the intent into the words themselves.
* **Stack sparingly.** Two tags before a clause (`[warm] [softly]`) normally combine well. Five stacked tags pull in different directions and will not produce a good result.
* **Unrecognized tags may be spoken.** If the model doesn't interpret a tag as direction, it can read the bracketed words aloud. Test any tag outside the tables above before shipping it.
* **The [voice](/tts/concepts/voices) has to support the direction.** A soft, narrative voice asked for `[shouting]` will only go so far. Pick a voice whose range matches the scene instead of forcing it with tags.
## Text formatting & punctuation
You can shape delivery without any tags, just by how you write the words. These conventions are language-agnostic and stack with audio tags — for example, `[sad] *please* don't go...` combines an emotion tag, an emphasized word, and a trailing pause.
### Formatting
| Convention | Effect |
| ------------------------- | ------------------------------------------ |
| `UPPERCASE` | Loud / shouting / strong emphasis |
| `*stress*` | Stress this word (pitch + emphasis) |
| `sooo`, `noooo`, `waaait` | Vowel elongation - drawl, sarcasm, whining |
| `I-I-I don't...` | Stutter / stammer |
| `lowercase everything` | Flat / muted / deadpan |
### Punctuation
| Convention | Effect |
| ----------- | ---------------------------------- |
| `...` | Pause / trailing off / hesitation |
| `—` / `--` | Abrupt cut-off / self-interruption |
| `,` | Short pause / breath |
| `!` / `!!!` | Rising excitement / intensity |
| `?` / `???` | Question / disbelief |
| `?!` | Incredulous / startled |
```text
You did WHAT?!
Oh, *you* noticed the haircut?
Wait... N-n-no. That can't be right.
```
Keep these conventions close to how people actually write. A few repeated
characters read as intent (`waaait`, `!!!`, `N-n-no`); dozens of them
(`whaaaaaaaaaaaaaaaaait`, `!!!!!!!!!!`) will degrade delivery. The same goes for stacking
everything at once — `SOOOO *AMAZING*!!!???` is more likely to sound off than
emphatic.
### Spoken interjections
Interjections the voice should actually say are written as ordinary words. **Do not** put these in tags.
| Sound | Conveys |
| ------------------- | ----------------------- |
| `Hmm`, `Hm` | Thinking |
| `Um`, `Uh` | Hesitation / fillers |
| `Mhm`, `Uh-huh` | Agreement |
| `Ugh`, `Eww` | Disgust |
| `Oof`, `Ow`, `Ouch` | Pain |
| `Mmm` | Pleasure / satisfaction |
| `Phew` | Relief |
```text
[clears throat] Hey, it's me. Got a sec? Mhm.
```
## Putting it together
The real power comes from layering both toolkits together. Below are three examples with fully annotated input. Pick a [voice](/tts/concepts/voices) whose character matches the scene and try in our playground.
```text
[whispering] Shh, shh — she's coming. [excited] SURPRISE!!! [loudly] Happy birthday!!!
[laughing] Look at her FACE! [happy] Oh, this is the best.
```
```text
[yawning] Okay... [drawn out] iiit's so late. [tired] [sighing] What a day.
[muttering] just five more minutes... [softly] g'night.
```
```text
[excited] Okay okay open it, open it!
[gasp] [awe] No way — is this the — [voice cracking] [happy] oh my gosh, I can't believe you remembered!
[sobs] [laughing] I'm crying, I'm actually crying.
```
Audio tags don't override the text. They work together with it. The model
reads the tag in the context of the words that follow, so `[excited]` on a
flat sentence sounds forced, while `[excited] SURPRISE!!!` sounds natural.
Write the words with the emotion already in them, then use a few well-placed
tags to lead the clause and let punctuation and formatting carry the rest.