Monday morning, meeting room, eight people, ninety minutes of strategy. Things get discussed, decided, discarded and decided again. At the end someone photographs the whiteboard, a colleague has jotted down bullet points, and three days later nobody remembers exactly who committed to the delivery date.
For online meetings this problem has long been solved: an assistant joins, transcribes and summarises. In the meeting room, however, many teams fail at one inconspicuous detail, the recording quality. This article is about why transcripts from in-person meetings disappoint so often, why better devices only partially change that, and which approach solves the problem at its root.
Why transcripts from the meeting room disappoint so often
Transcription models themselves are astonishingly good these days. When the result is still patchy, it is almost always the source material, meaning the recording. And in a meeting room, the recording faces three physical opponents:
- Distance. A microphone in the middle of the table captures the person next to it loud and clear. The voice from the far end arrives quieter, more reverberant and less distinct. The model has to guess, and guessing produces errors.
- Overlapping voices. In lively discussions people talk at the same time. On a single audio track those signals blend together, and no software in the world can fully separate them again.
- Speaker separation without an anchor. In online meetings every person delivers their own audio track, so attribution is unambiguous. In a room, a single device has to recognise who is speaking purely by sound. With similar voices, that leads to mix-ups in the minutes.
This explains an experience many people know: the same tool delivers clean results for video calls and, for the meeting room, a transcript you would rather not trust.
The usual fixes, and why they do not change the principle
The first reflex is better hardware. A conference microphone in the middle of the table, a dedicated recorder or a clip-on AI recorder. All of that helps a bit, because better microphones capture more. But it changes nothing about the principle: one device, one position, one room.
The currently popular AI recorders like Plaud or Pocket are subject to the same limit. They are strong for one-on-one talks and phone calls, but as soon as six people sit around a table, even the best single device records half the group from a distance. We examined this in detail in our comparisons of Plaud alternatives and the Pocket alternative. On top come the practical costs: buying devices per head, charging them, carrying them, not losing them.
The approach that sidesteps the physics: connecting several phones
The solution is not a better single microphone but more microphones, and they are already lying on the table. For the transcription of in-person meetings with Sally, several phones in the room can be connected. Each phone becomes the close microphone of the person it sits in front of, and Sally merges all audio tracks into one clean track.
That dissolves the three opponents from above:
- No more distance. Every voice is captured by a microphone right in front of the person, loud and clear, no matter how long the table is.
- Overlaps become untangleable. When two people speak at once, their voices sit on different tracks and stay distinguishable.
- Unambiguous speaker separation. Who is speaking is clear through the respective device instead of acoustically guessed. The result is speaker-separated and, according to Sally, reaches up to 98.8% accuracy.
And importantly for everyday use: a single phone is perfectly enough. Every additional connected device makes the result better, but nobody has to buy, charge or carry anything. The phones are there anyway.
How a transcribed in-person meeting works
In practice it is four steps. You open the Sally app and start the recording, with the phone lying visibly on the table. Other participants connect their phones to the same recording. After the appointment, Sally merges the tracks and creates the transcript, summary, decisions and tasks, exactly like for an online meeting. If you have fixed meeting rooms, you can additionally connect their room calendars, and Sally automatically knows when an appointment starts there.
One point belongs in every setup: consent. In Germany, Section 201 of the Criminal Code makes the secret recording of the non-public spoken word a criminal offence, and recordings are only allowed with everyone's agreement. Phones lying visibly on the table and a short announcement at the start create transparency here. Storage is GDPR-compliant and exclusively in Germany, details on the page about GDPR and security.
What this means for your team
The real payoff shows after the meeting. In-person appointments end up in the same place as online meetings from Google Meet, Zoom, Microsoft Teams and Webex, with the same summaries, the same search and the same task tracking. The whiteboard photo and the bullet-point sheet from the beginning of this article are replaced by a searchable record that states who committed to the delivery date, verbatim and with speaker attribution.
For teams that work on site a lot, at customers, on shop floors, in the meeting room, the documentation gap between digital and physical disappears. And all of it without a hardware budget, without device management and without anyone having to write minutes.
Conclusion
Transcribing in-person meetings rarely fails because of the AI and almost always because of the recording: one microphone for a whole room is a physical compromise. If you connect the phones that are already there instead, every voice is captured up close, speaker separation becomes unambiguous and the results match the level of online meetings.
Try it at your next real appointment: test Sally free for 30 days, or first take a look at how the transcription of in-person meetings works in detail.
