This is the paper for Voxtral Last week, Mistral released an open-sourced voice model that could rival paid voice AI. Voxtral comes in 24B and 3B versions, both run locally. Both use a Whisper encoder to turn raw waveforms into embeddings, a 4x downsampling adapt... See more

Transcribes and translates audio offline using Whisper https://t.co/EsSLkfuDZG