Speech Transcription

This page covers Mistral's Speech Transcription capabilities, including Offline and Realtime transcription. You'll learn how to integrate these features into your applications and understand their use cases.

Overview

Mistral's Speech Transcription services enable you to convert speech to text (STT) with high accuracy and low latency. We offer two main models tailored for different use cases:

Voxtral Mini Transcribe V2

Voxtral Mini Transcribe V2 is designed for batch transcription. It provides:

• High accuracy: Industry-leading transcription quality with low word error rates.
• Speaker diarization: Automatically identifies and labels different speakers in your audio.
• Context biasing: Allows you to guide the model with custom vocabulary for accurate transcription of domain-specific terms.
• Word-Level timestamps: Provides precise timestamps for each word, useful for subtitle generation and audio search.
• Multilingual support: Supports transcription in 13 languages, including English, Chinese, Hindi, Spanish, Arabic, French, Portuguese, Russian, German, Japanese, Korean, Italian, and Dutch.
• Noise robustness: Maintains high accuracy in challenging acoustic environments.
• Long audio support: Processes recordings up to 3 hours in a single request.

Voxtral Realtime

Voxtral Realtime is built for live applications. It offers:

• Ultra-Low latency: Configurable latency down to sub-200ms, ideal for voice agents and real-time applications.
• Streaming architecture: Transcribes audio as it arrives, enabling natural and responsive voice interactions.
• Multilingual support: Strong performance in 13 languages, ensuring global reach.
• Edge deployment: Can be deployed on edge devices for privacy-first applications with a 4B parameter footprint.
• Open weights: Available under the Apache 2.0 license on the Hugging Face Hub, offering flexibility and transparency.

Getting Started

Explore the features of Mistral's Speech Transcription services with our comprehensive guides:

• Offline Transcription: Learn how to use Voxtral Mini Transcribe V2 for batch transcription, including speaker diarization, context biasing, and word-level timestamps.
• Realtime Transcription: Learn how to integrate Voxtral Realtime for live transcription with ultra-low latency, perfect for voice agents and real-time applications.