What runs on your computer
Live speech recognition, audio-file transcription, positional speaker labeling, and meeting-note analysis run locally. Your working recordings and transcripts stay under your control on the PC rather than being sent to a browser service for each job.
Local processing can also make a voice workflow more predictable when a connection is slow or unavailable. Performance depends on the length and quality of the audio, the selected model, and the CPU, memory, and graphics hardware in the computer.
When an internet connection still matters
The application can connect during installation, speech-model downloads, license activation, updates, support navigation, and optional diagnostics. Those supporting tasks are different from sending each recording away for recognition.
If you work with sensitive material, combine local processing with sensible file handling: use an encrypted Windows account, understand your backup destination, restrict shared-folder access, and delete recordings or transcripts you no longer need.
Build a privacy routine around the whole file
A local transcript can still be copied into a cloud document, attached to an email, synchronized by a backup tool, or exposed through a shared Windows account. The processing location is only one part of the privacy picture.
Decide where recordings are saved, who can open them, how long to keep them, and whether generated notes contain information that should be redacted before sharing. The same care applies to meeting recordings, interviews, customer calls, and personal voice notes.