Glossary
Local speech model
A speech-to-text model stored and run on the computer, not on a remote server.
The model is the component that maps audio to text. Local means that component executes on the Mac. It can still mishear. Local is a statement about where the computation runs, not a statement that the transcript is perfect.
Murmur uses a local speech model. We do not name an architecture or a parameter count on this site, because the product description does not. History and stats stay in the app. The boundary of the claim is explained in What stays on your Mac when you dictate.