---
title: "Record and transcribe an in-person meeting — Memory Notch"
url: https://memorynotch.app/blog/record-in-person-meetings/
---

MEMORY NOTCH BLOG

# Record and transcribe an in-person meeting

Record and transcribe an in-person meeting: one device mid-table, consent first, and what Gemini, Copilot and the Granola app do with audio.

ON THIS PAGE

01Can Granola, Teams Copilot or Gemini transcribe an in-person meeting?02Laptop or phone: where should the device go in the room?03How do you tell speakers apart in an in-person transcript?04Do you need consent to record an in-person meeting?05How do you record and transcribe an in-person meeting for free?06What none of these tools does well07Where Memory Notch fits08Frequently asked questions09Sources10Related posts

[Open the live demo ↗](https://memorynotch.app/#live-demo)

Published November 6, 2026 · By [Cho Yin Yong](https://memorynotch.app/about/#cho-yin-yong)

To record and transcribe an in-person meeting, ask the room first. Then put one phone or laptop in the middle of the table and start an app that transcribes as it listens. On April 22, 2026, Google said Gemini's meeting notes would work in the room, so the big office suites now do this too. They differ in where the audio goes and whether the transcript can tell voices apart.

- **The setup:** one device, in the middle, closer to the quietest person than the loudest.
- **The order:** ask everyone, then press record. Recording on your own device doesn't excuse you from asking.
- **Who said what:** most tools number the voices. Names come from you, or from voice profiles: samples people recorded in advance so the software knows them.
- **The hard part:** two people talking at once. No tool separates overlapping voices well.
- **The choice that matters:** whether the audio stays on your device or goes to a provider's servers.

## Can Granola, Teams Copilot or Gemini transcribe an in-person meeting?

Yes, all three can, each with conditions. Gemini needs an eligible paid Google plan. Microsoft's two options both need a paid licence. Granola does it through its phone app.

**Gemini in Google Meet.** Google announced the change in a [Workspace post on April 22](https://workspace.google.com/blog/product-announcements/10-more-announcements-workspace-at-next-2026). Its note taker would now cover meetings "in-person, or hosted on another provider like Zoom or Teams." The [rollout notice of August 13](https://workspaceupdates.googleblog.com/2026/08/take-notes-with-me-for-in-person-meetings-is-now-available.html) explains how it works. You press "Take notes" on the Meet home screen, then press it again when the meeting ends.

Gemini then writes a summary, action items and the full transcript into a document in your Drive. The feature reached phones and the web between August 11 and August 31. The notice lists the eligible business, enterprise and personal plans. It doesn't mention speaker labels.

**Microsoft.** The Copilot phone app has a Record option under its plus button. Microsoft's [Record guide](https://support.microsoft.com/en-us/microsoft-365-copilot/get-started-with-record) says you need a Microsoft 365 Copilot licence on a work or school account. Each recording can run up to 120 minutes. The audio is saved to your work OneDrive, and you get a transcript and a summary in the chat.

For a meeting room, Microsoft sells Intelligent Speakers. Its [admin guide](https://learn.microsoft.com/en-us/microsoftteams/rooms/voice-recognition) says the room's transcript can name each person who has set up a voice profile. Those profiles live in your company's Microsoft account. It needs a Rooms Pro, Premium or Copilot licence. Microsoft suggests no more than 10 people in the room for the best accuracy.

**Granola.** The company's [transcription guide](https://docs.granola.ai/article/transcription) covers the phone app. It says the app "can also recognize different speakers during face-to-face meetings." It sends the audio to an outside transcription provider and doesn't save the audio itself. The phone app can't capture sound from other apps. For more, see [our Granola comparison](https://memorynotch.app/vs/granola/).

|  Tool |  How you start it |  What it needs |  Where the audio is processed |  Speakers in the room |
| --- | --- | --- | --- | --- |
|  Gemini in Google Meet |  "Take notes" on the Meet home screen |  An eligible Workspace or Google AI plan |  Google; notes saved to your Drive |  Not stated in Google's notice |
|  Copilot Record |  Plus button in the Copilot phone app |  Microsoft 365 Copilot licence, work or school account |  Microsoft; audio saved to your OneDrive |  Not stated in Microsoft's guide |
|  Teams Intelligent Speakers |  A Teams meeting in an equipped room |  Teams Rooms Pro, or Teams Premium or Copilot for the host |  Microsoft Cloud |  Names for enrolled voices, or numbers |
|  Granola mobile |  The Granola iPhone or Android app |  The Granola mobile app |  Granola's transcription provider |  Recognizes different speakers |
|  Memory Notch |  The app on a Mac, iPhone, iPad or Android phone |  No account |  On your device |  Numbered: Speaker 1, Speaker 2 |

Checked November 6, 2026.

If your company already pays for one of these suites, it's probably the approved choice. The notes also land where your colleagues already work. An on-device app fits when the conversation shouldn't reach a provider at all, or when there's no company account. That trade-off is the subject of [why your meeting notes don't need the cloud](https://memorynotch.app/blog/why-your-meeting-notes-dont-need-the-cloud/).

## Laptop or phone: where should the device go in the room?

Put the device in the middle of the table, as close as you can to the person who speaks most quietly. A phone usually wins, because it lies flat and fits in the middle. A laptop at the head of the table is close to one person and far from everyone else.

Distance costs more than you'd expect. A physics reference from Georgia State University puts it simply: ["doubling the distance drops the intensity by about 6 dB."](https://hyperphysics.gsu.edu/hbase/Acoustic/isprob2.html) That rule assumes open space. A real room bounces sound off its walls, so the drop is gentler. Still, the person two metres away reaches the microphone much more weakly than the person beside it.

| |  Phone |  Laptop |
| --- | --- | --- |
|  Placement |  Lies flat in the middle of the table |  Usually sits in front of one person |
|  Noise next to the mic |  Little, if nobody touches it |  Typing and trackpad clicks |
|  Battery |  Check it before a long meeting |  Plug it in if you can |
|  Best for |  Round tables, interviews, walking meetings |  Seminars where you also take notes |

A few habits help more than any setting:

- **Keep the microphone uncovered.** Lay the phone screen up, with nothing over its bottom edge.
- **Move it away from noise.** Air vents, projector fans and a cup set down beside the phone all end up in the transcript.
- **Don't type on the recording laptop.** A recent MacBook Pro has a "three-mic array," according to its [spec sheet](https://support.apple.com/en-us/121552). Those microphones sit in the same case as your keyboard.
- **Record a 30-second test.** Ask the person farthest away to say a sentence, then play it back before you start.

For a big room with one speaker at the front, the trade-offs change. [How to record a lecture](https://memorynotch.app/blog/how-to-record-a-lecture/) covers that case.

## How do you tell speakers apart in an in-person transcript?

The app groups stretches of speech by how each voice sounds and gives each group a number: Speaker 1, Speaker 2. This step is called speaker diarization. It doesn't know anyone's name. Names come from you after the meeting, or from voice profiles people set up in advance.

In a room, every voice reaches the same microphone, so the model has only the sound of each voice to go on. That works when people take turns. It gets harder when they interrupt each other, and harder again as the group grows.

On a hard public test set, NVIDIA's [own figures](https://huggingface.co/nvidia/Nemotron-3-Diarization) for its newest diarization model show 9.47% of speaking time labelled wrong with one to four people. With five to nine people, that rises to 28.65%. Our explainer on [what speaker diarization is](https://memorynotch.app/blog/what-is-speaker-diarization/) goes through why.

Three habits make the labels easier to fix afterwards:

1. **Go round the table at the start.** If each person says their name once, you know who each number is.
2. **Ask people not to talk over each other.** When two people speak at once, both voices end up in one mix, and the words can land on the wrong person.
3. **Rename after the meeting, not during it.** In a plain-text transcript, one find-and-replace turns "Speaker 2" into "Dana". [What speaker diarization is, and where it fails](https://memorynotch.app/blog/what-is-speaker-diarization/) explains why the labels are numbers.

Interview researchers face the same problem with stricter rules about anonymity. See [how to transcribe an interview for qualitative research](https://memorynotch.app/blog/transcribe-interviews-qualitative-research/).

## Do you need consent to record an in-person meeting?

Ask everyone before you start, every time. Whether the law requires their agreement depends on where you are, and your employer may require it anyway. Processing the audio on your own device changes who can read the recording. It doesn't change whether the people in the room agreed to it.

For the United States, the Digital Media Law Project, once run by a Harvard research centre, [explains the rules](https://www.dmlp.org/legal-guide/recording-phone-calls-and-conversations). It says federal law permits recording "in-person conversations with the consent of at least one of the parties." Some states require everyone's consent. The page lists ten, among them California, Florida and Pennsylvania.

The page also says it is no longer updated, so check your own state's current law. Elsewhere, look up your own country's rules.

Microsoft's guide says it plainly: "Get consent from everyone in the conversation before you start recording." It adds that its app doesn't tell the other people for you. That holds for every app here. A phone on the table looks like a phone, so say out loud that it is recording and what the transcript is for.

## How do you record and transcribe an in-person meeting for free?

You can, with limits. If you have a Pixel 6 or later, Google's Recorder app already transcribes and numbers speakers. Memory Notch lets you make short recordings free on every device it runs on.

**Recorder on a Pixel.** Google's [help page](https://support.google.com/pixelphone/answer/16269004?hl=en) says Recorder numbers the voices it hears on the Pixel 6 and later, and you can rename them afterwards. Speaker labels work only in American English. The voice analysis "happens on your phone or tablet."

**Memory Notch's free recordings.** Each free recording runs up to five minutes and keeps everything captured. That covers a stand-up or a quick hallway decision, not an hour-long review.

On the [iPhone](https://memorynotch.app/ios/) and [Android](https://memorynotch.app/android/) apps, your first two longer recordings run to the end. After that, a one-time purchase of $5.99 removes the limit. On a Mac, it's $79 once or $7.99 a month. All prices here are in American dollars. [How the five-minute trial works](https://memorynotch.app/blog/how-the-five-minute-trial-works/) has the details.

For phones that aren't Pixels, see [offline transcription apps for Android](https://memorynotch.app/blog/offline-transcription-app-android/).

## What none of these tools does well

- **Overlapping voices.** No app in this post claims to separate two people talking at once.
- **The far end of a long table.** A transcript can't recover speech the microphone didn't catch clearly. Move the device closer.
- **Big groups on one device.** Error rates climb past four voices, even on a model built for eight.
- **Names on their own.** Only the equipped meeting rooms put names on voices automatically, and only for people with a voice profile.

We didn't test Gemini, Copilot, the meeting-room speakers or Granola for this post. Their rows come from each vendor's own documentation.

## Where Memory Notch fits

Memory Notch records, transcribes and numbers speakers on the device you record with: a Mac, an iPhone or iPad, or an Android phone. There's no account to make. The phone apps make no network requests of their own, so the audio stays on the phone.

On the Mac, two NVIDIA models do the work: Nemotron for the live transcript and Nemotron 3 Diarization for speaker labels, with up to 8 speakers per channel. Recordings are plain files in `~/memory`. The Mac app makes three network requests, and none of them carries a recording; [the privacy policy](https://memorynotch.app/privacy/) lists them.

Memory Notch writes no automatic summaries, and speakers are numbered rather than named. Overlapping voices are hard for it too. The phone apps can't record phone calls or other apps' audio, so use them for meetings in the room or on speaker. Each device keeps its own recordings, and each app is bought separately. The [iPhone](https://memorynotch.app/ios/) and [Android](https://memorynotch.app/android/) pages list what each phone needs.

For meetings at work, see [recording work meetings on a Mac](https://memorynotch.app/work-meetings/). To try it, [download Memory Notch free](https://memorynotch.app/get/).

## Frequently asked questions

Can Granola transcribe in-person meetings?

Yes. Granola's transcription guide says its iPhone and Android app can recognize different speakers in face-to-face meetings. The phone app can't capture sound from other apps. Granola sends the audio to an outside transcription provider and doesn't keep the audio itself.

Can Teams or Copilot transcribe an in-person meeting?

Yes, with a paid licence. The Copilot phone app can record up to 120 minutes and returns a transcript and a summary, with the audio saved to your work OneDrive. In an equipped meeting room, the room's speaker can name people who have set up a voice profile.

Can Gemini take notes in an in-person meeting?

Yes. Since August 2026, Google Meet's note taker works in the room. You start it from the Meet home screen on the web or your phone, and Gemini writes a summary, action items and a transcript into a document in your Drive. It needs an eligible paid Google plan.

Should I record an in-person meeting on my phone or my laptop?

Usually a phone. It lies flat in the middle of the table, closer to everyone, and nobody types on it. A laptop works if it sits in the middle and you don't type on it while recording. Either way, record a short test with the person farthest away.

Is it legal to record an in-person meeting without telling anyone?

Whether it is legal depends on where you are. American federal law allows recording with one person's consent, but several states require everyone's agreement, and employers often have their own rules. Whatever the law says, ask the room first. Transcribing on your own device doesn't replace their consent.

How do I record and transcribe an in-person meeting for free?

On a Pixel 6 or later, Google's Recorder app transcribes and numbers speakers on the phone, with speaker labels in American English. Memory Notch's free recordings run up to five minutes on every device it supports. On its phone apps, your first two longer recordings run to the end.

## Sources

- [Google Workspace blog: 10 more announcements for Workspace at Google Cloud Next 2026 (April 22, 2026)](https://workspace.google.com/blog/product-announcements/10-more-announcements-workspace-at-next-2026)
- [Google Workspace Updates: Take Notes for me for in-person meetings is now available (August 13, 2026)](https://workspaceupdates.googleblog.com/2026/08/take-notes-with-me-for-in-person-meetings-is-now-available.html)
- [Microsoft Support: Get started with Record in Microsoft Copilot (checked November 6, 2026)](https://support.microsoft.com/en-us/microsoft-365-copilot/get-started-with-record)
- [Microsoft Learn: Voice recognition (voice profile) in Teams Rooms (updated 2026)](https://learn.microsoft.com/en-us/microsoftteams/rooms/voice-recognition)
- [Granola: How transcription works (checked November 6, 2026)](https://docs.granola.ai/article/transcription)
- [Google Pixel Phone Help: Use speaker labels in a recording (checked November 6, 2026)](https://support.google.com/pixelphone/answer/16269004?hl=en)
- [HyperPhysics, Georgia State University: Estimating sound levels with the inverse square law](https://hyperphysics.gsu.edu/hbase/Acoustic/isprob2.html)
- [Apple Support: MacBook Pro (14-inch, M4, 2024) tech specs](https://support.apple.com/en-us/121552)
- [NVIDIA: Nemotron-3-Diarization model card, Hugging Face (September 23, 2026)](https://huggingface.co/nvidia/Nemotron-3-Diarization)
- [Digital Media Law Project: Recording phone calls and conversations](https://www.dmlp.org/legal-guide/recording-phone-calls-and-conversations)
- [Memory Notch privacy policy](https://memorynotch.app/privacy/)

## Related posts

- [How to transcribe qualitative research interviews](https://memorynotch.app/blog/transcribe-interviews-qualitative-research/)
- [How to record a lecture on a MacBook or iPhone](https://memorynotch.app/blog/how-to-record-a-lecture/)
- [Offline transcription app for Android phones](https://memorynotch.app/blog/offline-transcription-app-android/)
- [What is speaker diarization, and where it fails](https://memorynotch.app/blog/what-is-speaker-diarization/)
- [Record work meetings on a Mac](https://memorynotch.app/work-meetings/)
- [Memory Notch for iPhone and iPad](https://memorynotch.app/ios/)

[Download Memory Notch](https://memorynotch.app/get/) · [More from the blog](https://memorynotch.app/blog/)