Skip to content

What to set up before the meeting starts

Up to date as of 14 November 2026

Most transcript problems start before anyone speaks. A tool can only work with the sound it's given, so a few minutes of setup beats an hour of fixing text afterwards.

What you need

  • A tool that's already been chosen and that you've checked the privacy terms of. Where recordings are stored explains what to look for.
  • A microphone close to the speakers. For a call, that's usually a headset or the device's own microphone. For a room, it's a speaker-phone or table microphone.
  • A quiet space with soft surfaces if you can find one.
  • The list of attendees with their names spelled the way you want them to appear.
  • A short list of words the tool may get wrong: product names, suburb names, staff names, acronyms.
  • An agenda, even a short one.
  • Permission from the people in the meeting. See the consent basics.

Fix the sound first

Speech recognition is hurt by the things around the speech. One source lists noise, echoes and the way the audio was recorded as things that distort speech1. It lists acoustic distortions such as echoes and room acoustics among the adverse conditions1.

You can't change the software, but you can change the room.

  • Close the door and windows. Turn off fans, heaters and anything that hums.
  • Pick a room with carpet or curtains over a bare one, if you have the choice, to cut echo.
  • Put the microphone nearer to people than to the air-conditioning vent or the laptop fan.
  • On calls, ask people to use a headset rather than speakers.
  • Ask people to mute when they're not talking.

Don't expect a perfect transcript even then. This just removes some common causes of a bad one.

Plan for how people talk

The same source says voices vary in accent, pronunciation, articulation, dialect, pitch, volume and speed1. It also says continuous speech, made of naturally spoken sentences, is harder to recognise1. In a meeting, everyone speaks naturally, so errors are normal.

What helps:

  • One person speaks at a time. Overlap is hard for any listener, human or software.
  • Ask people to say their name the first time they speak, especially in a group where voices may be new to the tool.
  • Say numbers, dates and names slowly, and repeat them. "The invoice is due on the fifteenth. That's the fifteenth."
  • Spell out unusual terms once, so they appear in the text.
  • Say decisions out loud and plainly: "So we've agreed that Priya will send the quote by Friday."

That last point does two jobs. It gives the transcript a clean line to pick up, and it makes action items and owners much easier to find later.

Tell the tool who's there

Many tools label speakers. That relies on diarisation, which is described as structuring the audio into speaker turns to improve the readability of a transcript2. It's used to answer "who spoke when?"2.

Whether your tool can match a voice to a name depends on the tool. Look for settings such as:

  • A field to enter or import attendee names.
  • A way to rename "Speaker A" to a real name after the meeting.
  • A custom word list or glossary for terms.
  • The language and regional variant, such as Australian or British English.

Check the language setting first. A tool set to the wrong English can mangle spelling and some words.

Settings to check before you press record

  • What's captured: audio, video, a transcript, or all of them.
  • Where it's saved, and who in your business can open it.
  • Whether a notice goes out to attendees automatically.
  • Whether the tool joins as a separate attendee, and who can see that.
  • Whether you can turn it off midway if the conversation turns sensitive.
Setup before you press recordTell people it's recordedFix room and microphoneAdd names and termsRun a short testStart the meeting
Do these in order, then run a quick test.

Run a short test

Record a minute. Talk the way you would in the meeting, with a few names and a number or two. Read the transcript back.

If a name keeps coming out wrong, add it to the word list or say it more slowly. If the voices are swapped, check the microphone position. If half the words are missing, the room or the connection is the problem.

Then start the real meeting. Afterwards, how AI meeting notes work explains why the checking still matters, and the next job is writing up the minutes.

Notes and sources

  1. Speech recognition, Wikipedia
  2. Speaker diarisation, Wikipedia