MEMORY NOTCH BLOG

How to transcribe a focus group with a seating map

Record the group on one device, get numbered speakers from the machine, then turn Speaker 3 into Seat 4 with a seating map and the note-taker's log.

Published November 6, 2026 · By Cho Yin Yong

To transcribe a focus group, record the whole session on one device in the middle of the table. Run it through a transcription app that labels speakers. Then match each numbered label to a seat, using a seating map and the note-taker's log. Last, check every quote you plan to use against the audio.

The machine numbers the voices. Only your notes can say whose they are.

Group size: Clemson University suggests 6 to 12 participants, plus a moderator and a note-taker.
Recording: one laptop on the table, a phone as backup.
Machine transcript: speakers come out numbered, never named.
Who is who: a seating map and the note-taker's log turn numbers into seats.
Checking: listen back to every quote you will use, and to any stretch where people talk over each other.
Cost: Rev's human transcription is $1.99 a minute on its own pricing page (checked November 6, 2026). A machine transcript costs only the app, plus your time checking it.

How do you transcribe a focus group?

You transcribe a focus group in four steps: record it cleanly, get a machine transcript with speaker labels, map the labels to participants, and check the transcript against the audio. Most of the work is in the third step, because a focus group has more voices than any other recording you will make.

Before and during the session:

Draw the seating map. Number each chair: Seat 1, Seat 2 and so on. Decide whether people will use pseudonyms in the room.
Start the recording before anyone talks, and keep it running through introductions. In the introductions each person speaks alone, in seat order.
Have the note-taker log first words. For each turn they catch, they write the seat and the first few words. "Seat 3: I stopped buying it" is enough to find the line later. After the session:
Get the machine transcript. A transcription app with speaker diarization (telling voices apart) gives you text split into numbered speakers.
Map the numbers to seats. Play the introductions, note which number each person got, then search the note-taker's first words in the transcript to confirm.
Replace the labels, with find-and-replace, from Speaker 2 to Seat 5 and so on.
Check against the audio. Read with the recording playing for any passage you will quote and any stretch where the labels jump around.

Clemson's guide gives the note-taker a broader job than this: the date, time, topic and participant numbers, quotes of special importance and body language. The first-words log adds one column to those notes, and it is the column you will use most at the desk.

Recording setup: one laptop or several devices?

One laptop in the middle of the table is the simplest setup that works, with a phone recording at the same time as a backup. Several devices spread around the room give you several files to line up by hand, and each file numbers the speakers in its own way.

Where to put it:

The centre of the table, at about the same distance from every chair. A laptop at the moderator's end hears the moderator loudest and the far chairs faintest.
On something soft. A folded napkin under a phone dulls the knock of cups and pens on the table.
Away from the air conditioner, the projector fan and the door.
Screen awake. On a Mac, keep the display from sleeping for the length of the session.

A second recording on a phone costs nothing to set up and covers the one failure you can't fix later: no audio at all. Put it at the other side of the table.

Memory Notch on a Mac records the built-in microphone as one channel and labels up to 8 speakers per channel. A session with 7 participants and a moderator fills those 8 slots. A group of 10 plus a moderator has 11 voices, so expect some people to share a number, and lean harder on the note-taker's log. Of the 6 to 12 people Clemson suggests, the smaller end of that range is also the end where the machine labels hold up best.

Telling participants apart

Speaker labels tell you that two stretches of audio sound like different people. They don't tell you which person sat in which chair, and they can't be trusted when people talk at the same time. The seating map and the note-taker's log are what turn labels into participants.

	Machine speaker labels	Seating map and note-taker's log
What it gives you	A number for every voice, on every line	A participant code for the turns the note-taker caught
Coverage	The whole recording	Only what one person could write down
Who is who	Not known; numbered, not named	Known, from the chair
Overlapping voices	Often merged into one label, or split oddly	Usually written as crosstalk, if at all
Effort	Minutes of processing	One person's full attention during the session

Used together, each covers the other's gap. The labels give the turn structure of the whole session, and the log gives a few anchor points per person to confirm each number. What speaker diarization is, and where it fails explains what a label claims, and what speaker diarization is covers how the model separates voices.

Overlapping voices are the hard part. Focus groups produce more of them than interviews, because agreement and laughter come in at once. When three people answer together, a transcript usually captures the loudest voice and drops or misattributes the rest. Mark those passages as crosstalk while you check, and don't build a finding on a quote from one.

Focus group transcript example

The lines below are an invented example, written to show the steps. No real participant said them, and the layout is not the exact file format of any app.

First, the machine transcript as it comes out:

Speaker 1: So when you last bought coffee for home, what made you pick that bag?
Speaker 3: Mostly the price. I stopped buying the nice one when it went up.
Speaker 2: Same. But I still get the nice one for weekends.
Speaker 4: I just buy whatever's on the end of the aisle.
Speaker 3: Right, the end of the aisle, yeah.

Then the same lines after mapping. The note-taker's log had "Seat 5: I stopped buying" and "Seat 2: still get the nice one".

Moderator: So when you last bought coffee for home, what made you pick that bag?
Seat 5: Mostly the price. I stopped buying the nice one when it went up.
Seat 2: Same. But I still get the nice one for weekends.
Seat 7: I just buy whatever's on the end of the aisle.
Seat 5: Right, the end of the aisle, yeah.

Two things to notice. Speaker 1 is the moderator, which you usually confirm from the first question of the session. And the log only had to catch two of the four voices; the rest came from the introductions.

Consent, anonymisation and storage for research

A research focus group needs each person's consent to be recorded. It also needs a plan for storing the recording, and usually pseudonyms in the transcript. Your ethics board sets the rules that apply to you. The points below come from two published university and research-data sources.

The review board at Columbia's Teachers College says researchers "cannot always guarantee their participants full confidentiality" in a focus group, because other participants may repeat what they heard. Its advice is in three parts:

Offer participants the option to be recorded.
Say in the consent form "how the recordings will be stored and used".
Pseudonyms are "an extra measure of privacy protection" that participants can use in the room.
Keep the master list that links names to codes "stored separately from the data".

The UK Data Service covers the transcript side. Its guide says to replace names and other direct identifiers with pseudonyms, or remove them, before sharing. Apply each replacement "consistently across transcripts and related materials". Keep an anonymisation log of the changes. It also warns that over-anonymisation "can remove too much contextual detail".

Where the transcription happens is part of the storage plan. A cloud service is one more place the recording goes, and your approval may need to name it. Transcribing on your own computer keeps the audio there.

Local transcription doesn't replace anyone's consent, or your institution's data plan. Why your meeting notes don't need the cloud covers that trade-off. For one-to-one interviews, coding and the same ethics questions in more depth, see transcribing interviews for qualitative research.

Cost: human transcription or machine?

A human service costs per minute of audio and does the checking for you. A machine transcript costs little or nothing per hour and leaves the checking to you. For a focus group, the label mapping falls to you either way unless the service knows who sat where.

Checked November 6, 2026.

	Rev, human transcription	Rev, AI plans	Memory Notch on a Mac
Price	$1.99 per minute	Essentials from $29.99 per seat per month, billed monthly	$79 once, or $7.99 a month
A 90-minute session	$179.10	Inside the monthly allowance (5,000 minutes per user on Essentials)	No extra cost
Turnaround	12 hours or less, per Rev	Not stated on the pricing page	Live, while you record
Accuracy claim	99%+, per Rev	Not stated on the pricing page	No accuracy figure claimed
Where the audio goes	Uploaded to Rev	Uploaded to Rev	Stays on the Mac

All prices here are in American dollars. The 90-minute figure is our arithmetic from Rev's per-minute price. Rev's strengths are real: a person listens to every minute, and the pricing page states a 99%+ accuracy figure and a turnaround time. For a funded study with a tight deadline and an approval that covers a vendor, that can be the right call.

What a machine transcript won't do

A machine transcript of a focus group gives you numbered speakers and text, and the rest is still your job. These are the limits of Memory Notch in particular:

Speakers are numbered, not named. Mapping Speaker 3 to Seat 5 is done by hand, with the seating map.
Up to 8 speakers per channel. The Mac's microphone is one channel, so a group with more than 8 voices will have some people sharing a label.
Crosstalk is hard to separate. Overlapping voices can be merged or misattributed.
No automatic summaries or coding. You get the transcript. Themes are yours to find, or to draft with an AI tool you choose, under that tool's data policy and your ethics approval. The AI tools guide shows how to hand it the files.

Where Memory Notch fits

Memory Notch records the microphone on a Mac, transcribes as people talk and labels speakers on the same computer. Each recording is saved as audio, a transcript and a Markdown note in ~/memory, so find-and-replace from Speaker 3 to Seat 5 works in any text editor. The Vocabulary tab takes brand and product names, along with what they get misheard as, which helps when a group discusses products. An Enhanced transcript re-transcribes a saved recording after the session.

The Mac app needs an Apple Silicon Mac with macOS 26 or later. Recordings up to five minutes are free; $79 once or $7.99 a month removes the limit. Download it and record a five-minute test in the room you'll use, before the session.

For the backup recording, the iPhone and iPad app and the Android app both record, transcribe and label speakers on the phone, offline, with no account. Both are free to download, and free recordings last up to five minutes; your first two longer recordings run to the end. A one-time $5.99 in-app purchase (the price in the United States) removes the limit, bought separately from the Mac licence. On Android phones with less than 8 gigabytes of memory, the transcript is made after you press Stop.

For any in-person meeting, not only a focus group, see how to record and transcribe an in-person meeting.

Frequently asked questions

How long does it take to transcribe a focus group by hand?

Typing a session from scratch is slow, which is why many researchers start from a machine or human transcript and correct it. The slow part of a focus group is attributing each line to the right person, and that work is the same whatever produced the first draft. A seating map and a note-taker's log of first words make it faster.

Can transcription software identify each focus group participant?

No. Speaker diarization tells voices apart and numbers them: Speaker 1, Speaker 2. It doesn't know who anyone is. You match each number to a participant using the introductions, a seating map and the note-taker's notes, then replace the labels with participant codes such as Seat 1 and Seat 2.

How many speakers can Memory Notch tell apart in one recording?

On a Mac, up to 8 speakers per channel, and an in-person recording uses the microphone, which is one channel. A focus group of 7 participants and a moderator fits. A larger group will have some people sharing a label, so the note-taker's log matters more.

Should a focus group transcript use real names?

Usually not. University ethics guidance commonly recommends pseudonyms or participant codes, with the list linking names to codes stored separately from the transcripts. The UK Data Service advises applying replacements consistently across all transcripts and keeping a log of what was changed. Follow your own ethics approval first.

Do I need consent to record a focus group?

Yes. Each participant should agree to be recorded and be told how the recording will be stored and used. The review board at Columbia's Teachers College notes that focus groups cannot guarantee full confidentiality, because other participants hear everything. Transcribing on your own computer limits where the audio goes, but it doesn't replace consent.

Sources

Download Memory Notch · More from the blog