Guide

How to record an in-person meeting on a Mac

Everything you know about recording calls on a Mac changes when the other people are in the room. There is no system audio to tap, nothing to diarize, and one hardware rule that silently ends more in-person recordings than any software bug.

Short answer
  • Recording a room is a microphone-only recording. There is no system-audio track, and no speaker separation either — speaker labels come from splitting the call’s audio, which does not exist in a room.
  • Closing a MacBook’s lid cuts the microphone in hardware, not in software. Apple’s platform security documentation says the disconnect prevents “any software—even with root or kernel privileges in macOS, and even the software on the T2 chip or other firmware—from engaging the microphone when the lid is closed.”
  • The built-in array is better than people expect at one metre and worse than they expect at three. Apple describes it as a “Studio-quality three-mic array with high signal-to-noise ratio and directional beamforming” — beamforming that is aimed at whoever is sitting at the keyboard.
  • Automatic stop degrades in a room. With no call audio to hear ending, Narada gives a mic-only session one fixed grace window past the scheduled end — about three minutes, or one if another meeting is already queued — instead of the speech-driven overtime a call gets.
  • Your Mac can stay awake with its screen off. Apple’s Battery setting “Prevent automatic sleeping on power adapter when the display is off” is described as “Prevent your Mac from sleeping when it’s plugged in and its display is off” — but the lid still has to be open.

Recording a meeting that is happening in the room is the simplest case technically and the easiest one to lose. Simplest, because there is no system audio to capture, no virtual audio device and no permission beyond the microphone — it is one input, and every recorder on your Mac can do it. Easiest to lose, because the failure modes are physical rather than digital: the lid, the distance across the table, the fan, the battery.

What is different about a room

On a callIn the room
Sources capturedMicrophone and system output, as two filesMicrophone only
Speaker separationThe call’s audio is split into Speaker 1, Speaker 2, …None — one track, everyone on it
Echo cancellationMeaningful: removes the call’s audio leaking back into your micNothing to cancel
Knowing when it endedThe app can hear the call go quietNo signal at all — a fixed grace window past the calendar end
What decides qualityThe codec at the other endWhere you put the Mac

The second row is the one people are most disappointed by, so here it is plainly: an in-person transcript from a Mac has no speaker labels. Labels come from diarizing the remote side’s audio, which arrives as a separate stream on a call. In a room, four voices and a chair scraping all arrive down the same microphone. Narada attributes that whole track to you, under the name in Settings, and does not guess further. If you need “who said what” from a room, this is the wrong tool and probably the wrong year.

The lid, which is not a settings problem

This is the number-one way an in-person recording is lost, and no amount of caffeinate, Amphetamine or “prevent sleep” tinkering will change it. Apple’s Platform Security guide documents a hardware microphone disconnect on Mac laptops:

"…prevents any software—even with root or kernel privileges in macOS, and even the software on the T2 chip or other firmware—from engaging the microphone when the lid is closed.”

It covers “All Apple silicon-based Mac laptops and Intel-based Mac laptops with the Apple T2 Security Chip”, the switch is driven by lid sensors, and Apple notes a delay of up to 500 ms before the microphone goes off. This is deliberate anti-spyware hardware. It is doing its job, and your recording is collateral.

So: lid open for the whole meeting. If someone shuts your laptop to make room for the pizza, the recording continues writing a file and the file contains silence from that moment. A good recorder will notice; see the last section.

The display, though, can sleep. Those are separate things. Apple’s Battery settings page lists “Prevent automatic sleeping on power adapter when the display is off”, described as “Prevent your Mac from sleeping when it’s plugged in and its display is off”. So the correct arrangement for a two-hour meeting is: plugged in, lid open, screen allowed to go dark. It looks less like you are staring at a laptop, which matters in a room.

If you genuinely need the lid shut — a laptop on a shelf, a Mac mini setup, a docked machine — use an external USB microphone. It is a different device and the lid sensor has nothing to do with it. Apple’s page says nothing about whether the internal array returns in clamshell mode with an external display attached, and I have not verified it either way, so do not plan a client interview around a guess. Test your exact hardware first, with a two-minute recording you actually play back.

Where to put the Mac

Apple specifies the MacBook Pro’s input as a “Studio-quality three-mic array with high signal-to-noise ratio and directional beamforming”. Both halves of that are true, and the second half is why the array flatters you and shortchanges the person opposite. Beamforming means the array prefers a direction — the person sitting at the keyboard. That is exactly right for a video call and exactly wrong for a table of six.

Then there is distance, which is not a software problem either. Direct sound falls off sharply as you move away from the source while the room’s reflections do not, so the ratio between the two gets worse the further away someone sits. At a metre you get a clean voice. At three metres across a hard table in a glass meeting room you get a quieter voice with the room smeared over it, and that is what the speech model has to work with.

Practical version:

  • Middle of the table, not in front of you. Give up your own advantage. You are the loudest source in the room by a wide margin and you do not need help.
  • Screen open and facing the quiet end — the beam follows the lid.
  • Off the table surface if people type or tap. Table knocks are conducted straight into the chassis and they clip. A folded jacket under the laptop is not audiophile nonsense, it is a decoupler.
  • Not next to the projector, the AC vent, or another laptop's fan. Continuous broadband noise is worse for transcription than one loud cough.
  • Nothing over the vents. A thermally throttled Mac spins fans, and the fans are centimetres from the microphones.
  • For anything bigger than four people, buy a cheap USB microphone. An omnidirectional desk or boundary mic in the middle of the table beats every built-in array, because it is in the middle of the table. Plug it in, select it as the input, and you have solved the hardest problem on this page with money instead of technique.

One expectation to reset while we are here. Published word error rates for on-device speech models — including the 4% to 37% range across languages that Narada’s transcription engine publishes for itself — come from clean benchmark corpora, not from a room with six people, an air conditioner and someone joining late. Far-field speech is harder than anything a benchmark measures. Placement buys you more accuracy than any setting in any app.

Power and heat for a long session

Plug in. Not because recording is expensive — capturing one microphone stream barely registers — but because the failure it prevents is total. Apple’s battery figures are quoted for video streaming and wireless web; nothing on a specs page tells you what two hours of capture plus on-device transcription costs on your machine, and a Mac that sleeps at minute 94 of a 100-minute meeting has produced a 94-minute recording at best.

The heavy work happens after the meeting anyway. Narada transcribes when the recording stops, and its on-device AI mode explicitly pauses while a recording is in progress — so the fans, if they come on at all, come on when the room is already empty. That is also why this is not a live-notes tool: nothing appears until the meeting has ended.

Stopping it, and why automatic stop is worse in a room

On a call, Narada can tell whether the meeting is actually over by listening to your Mac’s output: it keeps recording while there is still speech, polls every 30 seconds, stops after about three minutes of silence, and hard-caps an overrun at 60 minutes past the scheduled end. That whole mechanism needs an audio signal from the call. In the room there is none — the microphone hears the room whether the meeting is over or not, and “someone is still talking” is not a thing you can infer from a track that also contains chairs and corridor noise.

So a microphone-only session gets the degraded version: one fixed grace window past the calendar end — around three minutes, or about one if another meeting is already queued behind it — and then it stops. Consequences:

  1. A room meeting that runs 40 minutes over will be cut. Press +15 on the recording window before the grace expires, as many times as you need.
  2. A meeting with no calendar event has no end to work from at all. You start it from the menu bar and you stop it from the menu bar. There is no global hotkey — that is not built — so it is a click in the menu bar or the app window.
  3. Auto-record is not the shipping default. Out of the box Narada asks before each calendar meeting; automatic recording is something you turn on during onboarding or in Settings.

If the in-person meeting is on your calendar with attendees on it, everything else works as normal: the event’s title names the folder, the transcript and summary land under it, and the meeting joins the same archive as your calls.

The awkward part, in one sentence

On a video call, disclosure has a natural moment and a natural mechanism — half the tools in this category used to announce themselves by joining the participant list. In a room there is nothing but a laptop on the table that nobody is thinking about, and that asymmetry is the whole ethical problem with recording in person.

Say it at the start, before the small talk ends:

“I’m recording this on my laptop so I can write up the notes — say if you’d rather I didn’t.”

Two useful properties of that sentence: it explains the purpose, which is what people actually want to know, and it offers an exit, which is what makes it consent rather than notice. If someone says no, you take notes by hand. Narada cannot make a recording lawful, does not notify anyone on your behalf, and shows a reminder on your own screen and nowhere else. This is not legal advice and nobody selling software is qualified to give you any: rules vary by country and by state, and in some places the standard is that everybody agreed rather than that nobody objected — more on that here.

What you get afterwards, and what you do not

The recording becomes a folder: an .m4a, a transcript with timestamps as plain text, and — if an AI mode is on — a Markdown summary and a JSON analysis next to them. Files are unencrypted deliberately, so anything else on your Mac can read them. Transcription runs on the machine, with Parakeet TDT v3 through Core ML, across 25 European languages; the audio never leaves in any mode, and in Recorder-only or on-device AI mode nothing leaves at all. On-device AI is Apple Silicon only, with a one-time 3 GB or 7 GB model download depending on your RAM, and it produces lighter notes than the cloud route.

The honest limits, since a room recording sits closest to all of them:

  • No speaker labels, as above. One track, everyone on it.
  • No live transcription and no translation. Notes exist after the meeting, in the language spoken.
  • No iPhone app. If the conversation happens in a corridor, your Mac is not there and neither is Narada.
  • macOS audio can fail mid-recording, and no app can prevent that. A device reshaped underneath a live capture, an overloaded audio daemon, a microphone that quietly stops delivering — or a lid that closed. What Narada does is compare each finished recording against what was submitted to the file and mark it Incomplete, with a reason, in the list and on the meeting. It cannot recover audio that was never written. Silence in the middle of a room recording is unrecoverable in a way that a call is not: nobody else has a copy.

Which is the argument for a test recording before a meeting that matters. Thirty seconds, played back, listening for the far end of the table — it is the only check that tests the room, the placement and the hardware at once. If you already have a silent file, the silent-recording diagnostic works through the causes one at a time.

And for the once-a-quarter case, Apple’s own tools are free and on the machine: Notes will “Record audio sessions right in your note and generate live audio transcripts”, and Voice Memos is built for it. Both carry hardware and language gates — Notes transcription is “available on Mac computers with M1 or later” and only when the device language is one of ten (English, Spanish, Portuguese, Italian, French, German, Japanese, Korean, Simplified and Traditional Chinese), and Voice Memos transcription needs “macOS 15 or later and a Mac with Apple silicon”. Neither starts from your calendar or turns two hours into decisions and follow-ups — but for one lecture in one of those languages, the answer might simply be Notes.

Sources

  • Apple Platform Security, Hardware microphone disconnect — the lid disconnect, which Macs it covers, and the “root or kernel privileges” wording.
  • Apple, MacBook Pro technical specifications — “Studio-quality three-mic array with high signal-to-noise ratio and directional beamforming”, and the quoted battery figures (video streaming and wireless web).
  • Apple Support, Change battery settings on Mac — the “Prevent automatic sleeping on power adapter when the display is off” option and its wording.
  • Apple Support, Notes User Guide for Mac — “Record audio sessions right in your note and generate live audio transcripts”, and Record and transcribe audio in Notes — the M1-or-later requirement and the ten device languages. Voice Memos User Guide for Mac — the other built-in alternative.
  • Narada’s own behaviour — mic-only grace window, overtime rules, on-device transcription and the incompleteness verdict — from the shipping build and its source, version 1.2, macOS 15.6 minimum.

All checked 4 August 2026.

Questions people ask about this

Can a MacBook record audio with the lid closed?

No. Apple-silicon Mac laptops and Intel Mac laptops with the T2 chip physically disconnect the microphone when the lid closes, and Apple’s own documentation says no software can defeat it, “even with root or kernel privileges in macOS”. Keeping the Mac awake with caffeinate or a utility app does not help, because the Mac being awake was never the problem. An external USB microphone is a different device and the obvious workaround, but Apple’s page says nothing about clamshell mode, so test your exact setup before you rely on it.

Is the built-in MacBook microphone good enough for a conference room?

For a table of three or four people within about a metre and a half, usually yes. Across a six-person table with an air conditioner running, it degrades quickly — sound level falls with distance while room reverberation does not, so the far end arrives quieter and smeared. A cheap USB microphone placed in the middle of the table beats any built-in array in that room.

Does Narada label who said what in an in-person meeting?

No. Speaker separation runs on the system-audio track — the call’s audio — and an in-person recording has none. Everything in the room lands on the microphone track and is attributed to you, under the name set in Settings. If you need speakers separated, an in-person recording is not where you will get it.

Where does the recording end up?

In a folder you choose, one subfolder per meeting, holding an .m4a file plus a transcript with timestamps as plain text. If an AI mode is enabled there is a Markdown summary and a JSON analysis alongside them. Nothing is encrypted, on purpose, so other tools can read them.

Do I have to tell people in the room that I am recording?

In many places, yes, and in some the standard is that everyone agreed rather than that nobody objected. No app makes a recording lawful and no app tells the room for you — Narada shows a reminder on your own screen and nothing else. Say it in one sentence before you start.

Try it on your next call

Narada records your Mac's audio and your microphone, transcribes on-device, and hands you the decisions and follow-ups. Free to use, no account, no bot in the meeting.