Voxtral TTS 3
AI model · Text to speech · Text → Audio / speech by Mistral AI
Lifelike text-to-speech with zero-shot voice cloning for real-time agents.
Voxtral TTS 3 is a state-of-the-art text-to-speech model for creating lifelike voice agents with zero-shot voice cloning. It supports 35+ languages and low-latency streaming for real-time interactive applications.
Model specs
- Model family
- Voxtral TTS
- Version
- 3.0
- Released
- 2025-07-15
- Input
- Text
- Output
- Audio
- License
- Premier
- Capabilities
- Streaming
- Languages
- English, French, German, Spanish, Italian, Portuguese, Chinese, Japanese, Korean
Category: Generative AI
- Modalities
- Text → Audio
- Voice cloning
- Zero-shot and professional voice cloning supported
- Real-time streaming
- Low-latency streaming for real-time use cases
- Sizes and variants
- Voxtral TTS 3 (v3.0)
- License
- Premier (Commercial)
- Access
- Mistral API, Mistral Studio, and SDKs (Python, TypeScript)
- Zero-shot and professional voice cloning
- Supports 35+ languages and 50+ dialects
- Low-latency streaming for real-time agents
- Premier-tier commercial model (v3.0)
Overview
Voxtral TTS 3 is Mistral AI's state-of-the-art text-to-speech model designed for building lifelike voice agents. It provides zero-shot and professional voice cloning capabilities, allowing organizations to create unique and consistent brand voices. The model supports over 35 languages and 50 dialects, making it suitable for global deployments in customer support and media. Optimized for low-latency streaming, it enables real-time interactive applications where agents can speak and act naturally.
Voices and Languages
- Voice Cloning: The model supports zero-shot and professional voice cloning to create custom voices for agents.
- Multilingual Support: It supports more than 35 languages and 50 dialects for global speech generation.
- Lifelike Speech Generation: The model produces lifelike speech for voice agents and interactive applications.
Real-time Performance
- Real-time Streaming: The model features low-latency streaming for real-time agentic use cases.
- Agentic Integration: It is optimized for building agents that can listen, reason, speak, and act.
- Interactive Agents: Developers can talk with an agent from their terminal or browser in a few steps.
Use cases
- Developers build voice assistants that respond to users in real time with low-latency streaming.
- Organizations create custom brand voices using professional voice cloning for consistent customer interactions.
- Global companies deploy multilingual support agents that speak in over 50 different dialects.
- Media platforms generate natural-sounding narration for content across 35+ supported languages.
Specifications
- Modalities
- Text → Audio
- Voice cloning
- Zero-shot and professional voice cloning supported
- Real-time streaming
- Low-latency streaming for real-time use cases
- Sizes and variants
- Voxtral TTS 3 (v3.0)
- License
- Premier (Commercial)
- Access
- Mistral API, Mistral Studio, and SDKs (Python, TypeScript)
- Languages
- 35+ languages and 50+ dialects
- Version
- 3.0
- Dialects
- 50+
- Input Modality
- Text
- Output Modality
- Audio
- License Type
- Premier (Commercial)
- Streaming Support
- Yes
- Voice Cloning Type
- Zero-shot and Professional
How it works
- Create an API Key: Obtain a Mistral API key from the Studio console and store it in an environment variable.
- Install the SDK: Install the Mistral SDK with the voice-agents extra to pull in the necessary audio stack.
- Initialize Audio Session: Initialize an audio session using the SDK, providing instructions and selecting a voice ID.
- Run the Agent: Start the session to capture voice input and receive generated speech responses in real time.
FAQ
What is Voxtral TTS 3?
Voxtral TTS 3 is Mistral's state-of-the-art text-to-speech model with zero-shot and professional voice cloning, supporting 35+ languages and 50+ dialects.
Does it support real-time streaming?
Yes, the model supports streaming for real-time agentic use cases with low latency.
How do I change the voice of an agent?
You can change the agent's voice by using the tts parameter in the API, for example, setting voice_id to fr_marie_neutral.
Added by AntSpace from public sources (docs.mistral.ai).