shreyask/voxtral-pipeline-explorer

Voxtral Pipeline Explorer

An interactive visualization of the Voxtral Mini 4B Realtime speech-to-text pipeline.

Features

• —Pipeline Explorer — Click through each stage of the audio-to-text pipeline
• —Causal vs Bidirectional Attention — See how causal masking enables streaming
• —Delay–Accuracy Tradeoff — Adjust the transcription delay and see how it affects accuracy
• —Sliding Window Attention — Step through token-by-token attention with a bounded window

Learn More

• —Blog Post: From Sound to Text in Real-Time
• —Model Card
• —Mistral Blog: Voxtral Transcribe 2