Emotion and pace tags

Emotion and pace tags change how a voice says part of the text: write them inline in input, in square brackets, such as [emotion=happy] or [pace=slow]. To shape the delivery of the whole input instead, use style instructions.

Information

For how tags compare with the other ways to direct a voice, and how emotion works in Voxtral 2 and Voxtral 3, see Voices.

API reference — POST /v1/audio/speech

Tag syntax

Emotion and pace tags share the same syntax and rules:

Rule	Details
Format	`[emotion=<value>]` or `[pace=<value>]`, with a value from the tables on this page
Where	Inside `input`, before the text it applies to: at the start or in the middle of a sentence
Spoken	No: the tag isn't read aloud
Combine	Write an emotion tag and a pace tag one after the other, such as `[emotion=calm][pace=slow]`
Limit	You cannot use more than one tag of the same type on the same section of text
Scope	A tag applies from its position until the next tag or the end of the input.

Tag an emotion

Put the emotion tag in input, before the text to say with that emotion:

import base64
import os
from pathlib import Path

from mistralai.client import Mistral

client = Mistral(api_key=os.environ["MISTRAL_API_KEY"])

response = client.audio.speech.complete(
    model="voxtral-mini-tts-3-0",
    input="[emotion=happy]We passed the audit!",
    voice_id="voice_478",
    response_format="wav",
)

output = Path("happy.wav")
output.write_bytes(base64.b64decode(response.audio_data))
print(f"Saved {output.resolve()}")

import base64
import os
from pathlib import Path

from mistralai.client import Mistral

client = Mistral(api_key=os.environ["MISTRAL_API_KEY"])

response = client.audio.speech.complete(
    model="voxtral-mini-tts-3-0",
    input="[emotion=happy]We passed the audit!",
    voice_id="voice_478",
    response_format="wav",
)

output = Path("happy.wav")
output.write_bytes(base64.b64decode(response.audio_data))
print(f"Saved {output.resolve()}")

Supported emotions

The emotion tag takes one of these 34 values:

Emotion
`amused`
`angry`
`anxious`
`apologetic`
`appreciative`
`assertive`
`bored`
`calm`
`confident`
`confused`
`critical`
`curious`
`determined`
`disappointed`
`disgusted`
`excited`
`frustrated`
`grateful`
`happy`
`hopeful`
`playful`
`proud`
`reflective`
`resigned`
`sad`
`sarcastic`
`skeptical`
`somber`
`surprised`
`tense`
`tired`
`urgent`
`warm`
`wistful`

Pace tags

Pace tags set the speaking rate. Use them to slow down a disclaimer, or to speed through a list:

Pace	Effect
`very-slow`	The voice speaks much slower than its normal rate.
`slow`	The voice speaks slower than its normal rate.
`moderate`	The voice speaks at a moderate rate.
`fast`	The voice speaks faster than its normal rate.
`very-fast`	The voice speaks much faster than its normal rate.

The following example generates the same sentence at each pace:

import base64
import os
from pathlib import Path

from mistralai.client import Mistral

client = Mistral(api_key=os.environ["MISTRAL_API_KEY"])

for pace in ["very-slow", "slow", "moderate", "fast", "very-fast"]:
    response = client.audio.speech.complete(
        model="voxtral-mini-tts-3-0",
        input=f"[pace={pace}]Please read the terms carefully before you sign.",
        voice_id="voice_478",
        response_format="wav",
    )
    output = Path(f"pace_{pace}.wav")
    output.write_bytes(base64.b64decode(response.audio_data))
    print(f"Saved {output.resolve()}")

import base64
import os
from pathlib import Path

from mistralai.client import Mistral

client = Mistral(api_key=os.environ["MISTRAL_API_KEY"])

for pace in ["very-slow", "slow", "moderate", "fast", "very-fast"]:
    response = client.audio.speech.complete(
        model="voxtral-mini-tts-3-0",
        input=f"[pace={pace}]Please read the terms carefully before you sign.",
        voice_id="voice_478",
        response_format="wav",
    )
    output = Path(f"pace_{pace}.wav")
    output.write_bytes(base64.b64decode(response.audio_data))
    print(f"Saved {output.resolve()}")

Combine tags with a style instruction

Tags and a style instruction work at different levels. Style instructions are not available at launch. See Style instructions.

Unsupported values

The API doesn't reject a tag with a value outside the tables on this page. A request such as [emotion=ecstatic] returns a normal response, and the tag isn't spoken.

Disfluencies

Disfluencies add natural sounds such as pauses, breaths, and laughter to the speech. Write them inline in input, in square brackets, just like emotion and pace tags:

Disfluency	Usage
`um`	`[um]`
`uh`	`[uh]`
`laugh`	`[laugh]`
`breath`	`[breath]`
`sigh`	`[sigh]`
`gasp`	`[gasp]`
`sniff`	`[sniff]`
`cough`	`[cough]`
`throat clearing`	`[throat clearing]`
`yawn`	`[yawn]`
`chuckle`	`[chuckle]`
`laugh hard`	`[laugh hard]`
`gasping`	`[gasping]`
`rire`	`[rire]`

Information

This is a non-exhaustive list. Disfluencies use the same syntax as emotion and pace tags: [disfluency] inline in input.

Language support

Emotion, pace, and disfluency tags work in English. They are untested but expected to work in French, German, Spanish, Italian, Portuguese, and Hindi.

Going further