Back to Projects

Privacy-First Meeting Transcriber

AI/MLPython
Privacy-First Meeting Transcriber

Overview

A local, privacy-first meeting note-taker for Zoom, Teams, and Meet calls. Captures system audio without joining as a bot participant. Transcribes locally using faster-whisper (no cloud API calls for audio). Summarizes via GPT-4o to produce actionable meeting summaries with extracted action items. Optional speaker diarization and Google Calendar integration.

The Problem

Existing meeting transcription tools send audio to the cloud, raising privacy concerns. Many also require joining the call as a bot participant, which is intrusive and sometimes blocked.

My Approach

Captures system audio via BlackHole virtual audio driver — no bot joins the call. Transcribes locally using faster-whisper for complete privacy. Only the transcript text is sent to GPT-4o for summarization and action item extraction. CLI and Flask web interfaces for flexible usage. Optional pyannote.audio for speaker diarization.

Key Results

Local
Privacy
Audio never leaves machine
Whisper
Transcription
Local processing
Actionable
Output
Summary + action items

Tech Stack

Pythonfaster-whisperOpenAI GPT-4osounddeviceFlaskClick CLIBlackHole (virtual audio)