Voxtral Realtime Real-time speech transcription, entirely in your browser. Powered by Voxtral-Mini-4B

Load Model

This demo downloads and caches Voxtral-Mini-4B, a realtime transcription model optimized for in-browser inference (~2.8 GB).

2

Private & Local

Your audio is processed locally and never sent to a server. All inference runs on-device with Transformers.js and WebGPU.

3

Real-time Streaming

The model is capable of sub-500ms latency with support for 13 languages and a native streaming architecture.