Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.openwhispr.com/llms.txt

Use this file to discover all available pages before exploring further.

OpenWhispr detects when you’re in a meeting and offers to transcribe it in real time with speaker diarization.

​
Setup

Meeting transcription works out of the box — no calendar connection required. OpenWhispr detects meetings automatically through process and microphone monitoring.
1

Grant screen recording (macOS)

macOS requires screen recording permission to capture meeting audio. You’ll be prompted during onboarding.
2

Start a meeting

Meeting detection is on by default. When a meeting starts (Zoom, Teams, FaceTime, Google Meet), a notification asks if you want to record.
3

Optional: Connect Google Calendar

Go to Integrations > Google Calendar to auto-fill meeting titles and attendees in your notes. This enhances the experience but isn’t required for transcription to work.

​
How detection works

OpenWhispr uses two primary signals to detect meetings, with an optional third:
  • Process monitoring — detects Zoom, Teams, FaceTime, and Webex
  • Microphone activity — catches browser-based calls like Google Meet
  • Calendar awareness (optional) — when connected, shows the event name in the notification and auto-fills meeting details in the note
These signals are coalesced — you get one notification, not three.

​
Live transcription

Meeting audio is transcribed in real time via a cloud streaming provider (OpenAI Realtime, AssemblyAI Universal-3 Pro, or Deepgram). During a meeting:
  • Text appears as it’s spoken
  • Speaker labels are assigned live and refined after the call
  • A dedicated meeting hotkey lets you start/stop independently from dictation

​
Speaker diarization

OpenWhispr identifies who’s speaking during a meeting:
  • Live labels — assigned during recording
  • Post-processing — clusters are refined into stable speaker groups
  • Voice fingerprints — attach voice profiles to contacts so recognized speakers carry their name across meetings
  • One-click reassignment — fix any mislabeled speakers after the call

​
Controls

A floating pill appears at the top of the transcript during recording. From there you can:
  • Toggle identification off for this meeting — transcripts fall back to “You” / “Others” labels based on audio source (your mic vs system audio). Your global setting stays unchanged.
  • Adjust expected speaker count — a “1 other in call” stepper. Calendar attendees pre-fill the count; bump it up if more people join. Unscoped recordings cap at a sensible default to prevent phantom speakers.
To turn diarization off by default across all recordings, go to Settings > Meetings > Identify and label speakers.

​
After the meeting

Meeting transcriptions are saved as notes. You can:
  • Review and edit in the note editor
  • Apply AI actions to clean up or summarize
  • Search across past meetings using semantic search

​
Echo cancellation

OpenWhispr includes a WebRTC AEC3 sidecar that removes mic echo from captured system audio. If the native helper isn’t available, a JavaScript fallback handles it.