How to Transcribe Zoom, Teams, and Meet Locally

Why Real-Time Transcription Matters

Taking notes during a meeting splits your attention. You miss parts of the conversation or end up with useless scribbles. Real-time transcription solves it — words appear on screen as they're spoken, and you stay present.

Audio Note has three real-time modes: Microphone, App Audio (a specific app such as Zoom or Teams), and System Audio (everything the computer is playing). It runs on Mac and Windows.

Setting Up for a Meeting

Before the Meeting

1. Open Audio Note and select Realtime from the sidebar
2. Choose your audio source:
   • For online meetings (Zoom, Google Meet, Tencent Meeting, etc.): select App Audio and pick the meeting app from the dropdown
   • For in-person meetings: select Microphone
3. Pick a transcription model. The "Streaming" model is optimized for low latency — it adds words as they're spoken rather than waiting for full sentences
4. Click Start

If you need to evaluate latency, test with a short representative meeting clip on your own setup. The experience depends on the machine, model, and audio source.

Zoom, Teams, and Google Meet

Cloud meeting bots join the call as a participant. Local capture does not. You transcribe what your computer already plays, so the other people in the meeting do not see a notetaker join.

For Zoom, Microsoft Teams, Google Meet, or similar apps:

1. Start the meeting in its own app or browser
2. In Audio Note, choose App Audio and pick that app. If the app is missing from the list, use System Audio
3. Start transcription before the discussion you care about. There is no bot waiting room to admit
4. Keep the meeting window unmuted on your side. Local capture records playback, not the other person's upload path

App audio is usually cleaner than system audio because it ignores music or notifications from other programs. System audio is the fallback when the meeting runs in a browser that does not appear as its own app.

Permissions differ by OS and meeting app. Do a two-minute test call before a real meeting. If the transcript stays empty, the OS has not granted audio capture yet.

During the Meeting

• The transcript scrolls automatically. You can pause it anytime to review a section
• Use the flag button to mark important moments — these become bookmarks in the final transcript
• Speaker labels appear automatically if the model detects different voices
• You can type inline notes alongside the transcript without interrupting the recording

After the Meeting

Stop the recording. Audio Note saves the full transcript with timestamps and bookmarks. From here:

1. Click AI Chat and select Meeting Notes — it formats the transcript into structured minutes with attendees, discussion points, and decisions
2. Select Action Items to extract tasks with owner detection
3. Export as Markdown to share with attendees who missed the meeting

Tips for Better Results

Audio Quality Matters Most

Real-time transcription quality depends more on the input audio than the model. A few things that make a noticeable difference:

• Use an external mic for in-person meetings. The built-in laptop mic can work in a pinch, but external hardware often gives a more stable input
• Keep the mic close to primary speakers and away from air conditioning, keyboard noise, and loudspeakers
• In a noisy environment, enable the noise filter in Audio Note's settings

Speaker Separation

Speaker separation is easier to review in small, quiet meetings. For larger groups or noisy environments, results are less reliable. A practical workaround: assign speakers manually after transcription using the editing tools.

Meeting Length

There's no hard limit on session length. Longer meetings produce longer transcripts, so editing, search, and AI cleanup depend more on your machine and project size. For important meetings, split recordings by agenda section or time block to make review easier.

A Note on Privacy

All real-time processing happens on your computer. Audio Note runs Whisper or Sherpa-ncnn locally, and the transcript is stored in the app's local database. The audio is not sent to a transcription server. That is the point of using this instead of a cloud meeting bot.