How do you transcribe a one-hour meeting with several speakers?
The short answer
Record the room on a single device placed at the centre of the table, then run the audio through a tool that separates the speakers on the sound itself before transcribing, rather than one that transcribes a flat file and guesses who spoke afterwards. That separation step, called diarisation, is what decides whether a one-hour meeting with five or six people comes back readable or comes back as an unusable wall of text.
Generic voice dictation and video-call tools fail here for a structural reason: they are built for one microphone per person, or for one person speaking close to the microphone. A meeting room is the opposite situation, with distant voices, overlapping speech and background noise.
Meetingize is built for exactly that case, handles up to 3 hours and 20 speakers, and returns a transcript attributed speaker by speaker plus the finished write-up. The free plan is there so you can test it on your own difficult recording rather than take our word for it.
Why multi-speaker meeting transcription is genuinely hard
Transcribing one person speaking into a headset has been a solved problem for years. Transcribing a meeting is a different exercise, and the difficulty is in the room rather than in the language.
Three things go wrong at once, and they compound each other. Understanding them is what lets you tell a tool that will work from one that will waste your afternoon.
- Distance. A voice five or ten metres from the microphone arrives quieter than the air conditioning and the chairs. The words are physically there, buried under room tone, and a tool that does not separate speech from noise before transcribing simply drops them.
- Overlap. In a real meeting people interrupt each other, finish each other's sentences, and talk over the person chairing. A system that assigns speakers by reading the text afterwards has no way to resolve two voices occupying the same second.
- Attribution over time. Over an hour, a voice has to stay the same voice. If the attribution drifts halfway through, the transcript stops being a record of who committed to what, which is usually the only reason you wanted it.
The practical setup before you press record
Most of the quality is decided in the first thirty seconds, and none of it requires equipment. Whatever tool you use afterwards, these are worth doing.
- One device at the centre of the table rather than in front of the chair. A phone lying flat in the middle of a table of six beats a laptop at one end.
- Nothing on top of the microphone. A notebook, a jacket or a hand over a phone costs more than any software setting will give back.
- Tell the room it is being recorded, at the start, out loud. It is required practice in most professional settings, and it also gets each person to say their name once, which helps you map the speakers afterwards.
- Do not split the meeting into five separate recordings. One continuous file lets the tool keep the same speaker identity from beginning to end.
What a proper multi-speaker transcription chain actually does
The difference between tools is not the transcription model, which most of the market buys from the same handful of suppliers. The difference is the chain around it. In Meetingize, an hour of meeting goes through these stages.
- Speech is isolated from the rest of the room, so the transcription only ever works on actual speech instead of drowning in silence and noise.
- Speakers are separated on the acoustics, as their own stage, before anything is transcribed. This is the step that holds up when someone is far away or when two people overlap.
- Transcription then runs with the speaker map already in hand, so sentences are cut at real speaker boundaries rather than in arbitrary places.
- The speaker map and the text are merged by our own processing, with duplicated fragments at segment boundaries removed and turns aggregated per speaker. You get a dialogue, not a stream of snippets.
- Meetings held in another language, or mixing languages mid-sentence, stay within the same chain, and the write-up can be produced in the language you need.
From transcript to a meeting report you can actually send
A one-hour transcript is roughly nine thousand words. Nobody reads that, including you, which is why a raw transcript is not really the deliverable you were after.
What you needed was the automatic meeting report: the decisions, the commitments with names attached, the open points, and in a regulated profession the document that the meeting existed to produce. Meetingize generates that document from the attributed transcript, in maintained professional templates, and opens it in an editor with the rendered result beside the source so you can correct a name and export to PDF or Word.
If you took photographs during the meeting, of a whiteboard or a document, they are described and placed in the report at the right point rather than left as a separate attachment.
What the available options are really worth on this case
There is no single right answer, it depends on whether your meetings happen in a room or on a video call. Here is what each family of tools does well and where it stops, including ours.
A dictaphone or your phone's voice recorder, then typing it up
What it does well: Free, always available, and the recording itself is usually fine. As a raw archive of what was said, a phone in the middle of the table does the job.
Where it stops: There is no transcription and no attribution, so the hour of meeting becomes two or three hours of listening back and typing. In practice the file is never reopened, and the meeting ends up documented from memory a day later.
Built-in transcription in Teams, Zoom or Google Meet
What it does well: Excellent when the meeting really is a video call, because each participant has their own microphone and their own audio stream. Attribution is then close to free, and it costs you nothing extra.
Where it stops: It collapses when people share a room, because several voices arrive on one stream and the per-participant advantage disappears. It also stops at a transcript and a recap: no professional document at the end, and no regulated templates.
A dedicated recorder such as Plaud
What it does well: Purpose-built hardware, comfortable to use, and well reviewed by the people who own one. For capturing a room, it does what it says.
Where it stops: The recording was never the hard part. What comes back on a long multi-speaker meeting is a general-purpose summary, which flattens a two-hour discussion into bullet points. Their legal notices also name a Hong Kong entity and their storage is stated in the United States, which is a conversation worth having if the audio is a client meeting under professional secrecy.
Otter, Fireflies and the video-call assistants
What it does well: The reference tools for online meetings, where a bot joins the call and records each participant separately. They are better than us at that specific job, and we say so plainly.
Where it stops: In a physical room they fall back to a single microphone, which is exactly the hard case. They also stop at transcript, summary and action items, and they state no EU data residency on their standard plans.
Meetingize
What it does well: Built for the meeting in a room: a chain that separates speakers on the acoustics before transcribing, up to 3 hours and 20 speakers, then the finished professional document in maintained templates. Hosted in France, operated by a French company, under GDPR.
Where it stops: No bot joins your Zoom or Teams call in your place: you record the meeting or you import the audio afterwards. If all your meetings are video calls and a transcript is all you need, an assistant that joins the call is the more sensible choice.
Questions people ask about transcribing a multi-speaker meeting
How many speakers can be transcribed in the same meeting?
Meetingize handles up to 20 speakers in a single recording, for meetings up to 3 hours. The speakers are separated on the audio itself, so the limit is about the recording rather than about how many names you can type in.
Do I need a conference microphone, or is a phone enough?
A phone placed flat at the centre of the table is enough for the large majority of meetings, and it is what most of our users do. A conference microphone helps in a big or echoey room, but it is an improvement rather than a requirement.
How long does it take to transcribe a one-hour meeting?
Processing runs in the background and you are notified when the documents are ready, so you do not wait in front of a progress bar. In practice you start the processing as you leave the room and read the write-up rather than write it.
Can I import a recording I made with another device?
Yes. Upload an audio file recorded with a dictaphone, a dedicated recorder or your phone, and it goes through the same chain as a recording made in the app: speaker separation, transcript, then the structured document.
Does the transcript say who said what?
Yes, the transcript is attributed turn by turn, and you can rename the detected speakers with the real names once. That attribution is what the generated document relies on to state who committed to what.
Where is the audio stored, and is it used to train models?
Recordings, transcripts and documents are stored on servers located in France, operated by a French company, under GDPR. Your content is not used to train models, and personal data is replaced by reversible tokens before any text is sent to a language model.
Test it on the worst recording you own
Take a long meeting, in a room, with people far from the microphone, the one no tool has handled properly so far. Run it through Meetingize and read the write-up. The free plan exists for that comparison.
Commencer gratuitement