Speech is turned into text on the Mac itself. No audio leaves the machine.
What works today
- Local speech models on Apple’s MLX: Qwen3-ASR and Whisper large-v3.
- Translation from any supported language into English.
- A local server that speaks the standard audio interface, so scripts and agents can transcribe files with the usual client library.
Why it is in the lab
It is the daily-use proof that a private, local model can replace a hosted service for one real task. It also gives speech-to-text to the call-analytics stack.