VoiceTTS

Mistral

Text-to-speech

View as MarkdownOpen in Claude

Mistral’s Voxtral text-to-speech models deliver realistic, emotionally expressive multilingual speech with low latency.

Models

ModelDescription
voxtral-mini-tts-latestDefault. Latest Voxtral Mini TTS model

Model IDs are passed through to Mistral directly, so newer Voxtral TTS models work as soon as Mistral releases them. See Mistral’s text-to-speech documentation for current models.

Voices

Press play to audition any Mistral voice, then copy config to grab the value for SWML or your SDK.

Voice IDs are passed through to Mistral’s hosted voice library — a voice named myvoice becomes mistral.myvoice in the voice string. Find the available voices in Mistral AI Studio under Audio › Text-to-speech.

There is no default or fallback voice — always set a valid Mistral voice ID. Note that the preset names published with the open-weight Voxtral model (such as casual_male) are not accepted by Mistral’s hosted API — use a voice from Mistral AI Studio.

Learn more in Mistral’s documentation:

Mistral TTS docs

Reference for Voxtral TTS models and voices.

Voxtral TTS announcement

Overview of Voxtral TTS capabilities.

Languages

Voxtral TTS supports English, French, Spanish, Portuguese, Italian, Dutch, German, Hindi, and Arabic. Language is determined by the model and voice; refer to Mistral’s documentation for the most up-to-date language coverage.

Usage

A voice identifier string has three parts: the mistral engine code, a voice ID, and an optional model.

Format: mistral.<voiceId>:<model>

Examples (replace myvoice with a voice ID from Mistral AI Studio):

mistral.myvoice
mistral.myvoice:voxtral-mini-tts-latest

You can also set the model with the separate model parameter of the languages SWML method instead of appending :model to the voice string:

1languages:
2- name: English
3  code: en-US
4  voice: mistral.myvoice
5  model: voxtral-mini-tts-latest

Build with Mistral on SignalWire

1

Create a Space and add credit

If you don’t have one yet, you’ll need to create a SignalWire Space. Be sure to add some credit to test with.

2

Add a new Resource

Navigate to the Resources tab in your SignalWire Dashboard and click + Add New to create a new Resource.

3

Create a SWML Script

From the Resources menu, select SWML Script. Name it something fun and recognizable. Ours is titled Mistral Wizard.

Next, paste the following starter script into the text box, replace myvoice with a voice ID from Mistral AI Studio, and hit Save:

1version: 1.0.0
2sections:
3  main:
4  - ai:
5      prompt:
6        text: |
7          You're a voice from Mistral's Voxtral TTS engine!
8          Introduce yourself, and have a conversation about programmable unified communications on the SignalWire platform.
9      languages:
10      - name: English
11        code: en-US
12        voice: mistral.myvoice
4

Buy and assign a phone number

Navigate to the Phone Numbers section of the Dashboard’s left sidebar menu.

Purchase a phone number and assign it to the desired SWML script.

A purchased phone number showing assignment to a specified Resource.

Assigning a phone number to the SWML Script
5

Give it a call!

Call the number you just assigned to chat with your new AI voice application on the phone.

Next steps with SWML

Now you’ve deployed your very first SignalWire voice AI application using Mistral voices. Next, dive deeper into SWML to explore its capabilities!

Methods reference

Documentation for all SWML methods

AI in SWML

Build advanced AI applications using SignalWire Markup Language

Guides

SWML guides and demo applications