Your Meeting Transcript Has Every Word, but Not Who Said Them

What You’ll Learn

• Why a transcript can be complete, accurate, and still not tell you who spoke

• What one shared microphone does to a public record, measured across a single meeting

• Why the fix isn’t more accuracy but a way to tell voices apart

• Why a name that does survive might not survive intact

• Four things to check in your own transcript, in about twenty minutes

Try This Before You Read Any Further

Open your most recent meeting transcript in another tab and find any vote.

Now scroll up a few lines and see if you can tell who proposed the item being voted on, and who backed them up. In meeting language one person moved it and another seconded it, and both names are supposed to be part of the record.

If they’re both right there, you’re in good shape, and the rest of this piece won’t apply to you. If you worked it out from something the clerk said a few lines earlier, keep that in mind. And if you scrolled up and down twice and came away with nothing, you’ve just found the gap this piece is about, and so have most councils.

A City That Put the Problem in Writing

We read through a full council meeting transcript from a mid-sized American city, published by the city itself, the way any resident could. Before you reach a single word of the meeting, there’s a note from the city.

It explains that the council was speaking through a shared room microphone, and that as a result the transcript cannot say who is speaking. Then it says the city knows about the issue and is working on a solution.

That’s not a disclaimer buried in a footer. It’s the first thing on the page, above everything. Somebody looked at their own record, worked out what was wrong with it, and put the finding on top of the evidence.

So this isn’t a story about a city getting caught. It’s a story about what the rest of that transcript shows, which is a problem plenty of councils have and far fewer have written down.

We’re not naming the city, and we’re not linking the file. The point isn’t which council this was. It’s that you can probably find the same gap in yours.

One Microphone, One Name, a Whole Council

Here’s the shape of it. Every turn in that transcript carries a speaker label, and for most of them the label is the name of the room rather than the name of a person. Count them and you get 843 turns attributed to the room, against 52 turns spread across six named individuals.

It’s a whole council sharing one microphone and one name, the most collaborative byline in local government.

The six people who do get named have one thing in common, and it isn’t seniority or how much they talked. Every one of them was on their own audio channel. The chair introduces four of them as joining virtually, a fifth is described as being on the line, and the sixth is the staff member running the virtual queue. The council members in the room were sharing one microphone. That’s the whole difference: not budget, not vendor, not the year the system went in, but whether the system had any signal it could use to tell one voice from another. In that room, the only signal it had was which microphone you spoke into.

That matters more than it sounds, because it rules out the explanation you’d reach for first. Compare two cities and one has better records, and you can always say the other one had more money. Here there’s nothing to compare, because it’s the same room, the same night, the same minute of the same meeting.

And the document runs the experiment for you, which is more than most sources do.

Two city staff experts speak at this meeting. One is in the room, and he opens by stating his name and job title for the record, exactly the way you’re taught to. Every turn he takes is still attributed to the room. The other joins on the line, says “Hi, guys,” volunteers nothing about who he is, and gets his own name on every turn.

It was the same meeting, the same employer and the same kind of contribution. One of them announced himself and the record forgot him anyway. The other said nothing about who he was and the record knew.

What Goes Missing: Motions, Seconds and Votes

Fair question at this point: is this a real problem or a tidiness problem? Watch a vote.

The council takes up a motion. Here is the shape of it on the page. This is the structure, not a quotation:

[room]: So moved.

[room]: Second.

[room]: Motion and a second. All those in favor.

[room]: That’s unanimous.

The motion carried and the vote is in the record, but who moved it and who seconded it can’t be recovered from that document by careful reading, by context, or by anyone, and it’s gone from the public record.

Now the roll call, where it very nearly works:

[room]: Roll call begins. The clerk reads a councilmember’s name.

[room]: Yes.

Look at what’s actually holding that record together. The “Yes” is attributed the same way as all 842 others. You can work out whose it is, but only because a human said a name out loud a second earlier, and only for as long as the names keep coming in order. When two answers come at once, which happens in this document, the mapping breaks.

Your attribution isn’t a feature of your recording. It’s a side effect of one person reading a list without losing their place.

The Same Name, Spelled Four Ways

Here’s where the document gets uncomfortable.

The same transcript spells the same people two different ways.

One participant appears six times under one spelling of her first name and three times under another, in the same meeting, about the same person. Three council members are each rendered two ways. So is the fire department official who joined on the line: eleven turns one way, one mention the other. That’s the version you’d never catch, because eleven-to-one doesn’t look like an error. It looks like a typo you can ignore.

The worst case is a member of the public. Her name reaches the record four different ways in one meeting: one spelling on every turn she speaks, two different manglings when the chair calls her up, one of which splits her surname into two separate names, and a fourth version when she introduces herself. That’s four renderings of one person inside twenty minutes.

That’s why there isn’t a single name in this piece, and why we haven’t told you which city it was. Printing a real person’s name in a piece about records that get names wrong felt like exactly the wrong place to make that mistake ourselves.

So the failure isn’t only that the record can’t say who spoke. Where a name does survive, it doesn’t reliably survive intact. A resident searching your archive for what their council member said won’t find some of it, and won’t know they didn’t.

Why “More Accurate” Doesn’t Fix It

The instinct is to treat this as an accuracy problem and look for a better transcription engine. That’s solving a different problem.

Accuracy is about whether the words are right, and in this transcript the words are largely fine. What’s missing is attribution, and attribution doesn’t come from listening harder. It comes from giving the system a way to tell one voice from another. A separate microphone channel per person is one way to do that, and it’s a crude one: it tells the system where the sound came from, not who made it, and it stops working the moment two people share a mic. Recognizing the voice itself is a better one, because the signal travels with the person instead of with the equipment.

That’s why the six named people in that transcript are named. They each arrived on their own channel, which happened to be the only signal this system knew how to use. Everyone who shared the room microphone was invisible to it, no matter how clearly they introduced themselves.

It’s also why this reaches you as an accessibility issue when it started as a recordkeeping one. The Americans with Disabilities Act (ADA) Title II web rule sets Web Content Accessibility Guidelines (WCAG 2.1 Level AA) as the standard for local government web content, and WCAG’s own definition of captions lists speaker identification among the things a caption is supposed to convey, because knowing who’s talking is part of what makes a caption usable for a viewer who is deaf or hard of hearing. A transcript that collapses a room into one label has exactly the same hole. It’s the same cause, noticed by two different audiences, probably in two different emails to you.

Four Things to Check in Your Own Transcript

None of this needs a project. You’ve already got the document open from earlier. Look for four things.

Do your speaker labels name people, or name the room? If one label is doing most of the work, you’ve got the same pattern.

Pick a motion. Can you tell who moved and who seconded? If the answer is “only because somebody said it out loud,” your attribution depends on procedure rather than on the record.

Search one council member’s surname and count the spellings. More than one is the quiet version of this problem, and it’s the one nobody reports because nobody notices.

Ask who was on their own microphone. That list is usually the same as the list of people your transcript can name. If it is, your system is telling voices apart by channel, and everyone who shares a channel is invisible to it.

That’s twenty minutes on a document you already have. Whatever you find, you’ll know something about your own record that you didn’t know this morning.

And if what you find looks like the transcript we read, take the rest of that city’s example too. They looked, they wrote down what was wrong, and they put it where residents could read it instead of quietly hoping nobody would check. The engineering part has a solution. Saying it out loud is the harder part.

MediaScribe Live identifies speakers by their voices rather than by which microphone they spoke into. A council member’s voice is learned once and recognized at every meeting after that, on the microphones already in the room, with no new mics to buy [and without storing anyone’s actual voice recording]. What that changes is the part of your week that goes into cleanup: less time fixing attribution by hand after every meeting, and a transcript residents can search and trust. If your transcript looks like the one we read, book a call with the MediaScribe team and bring the transcript.

FAQ

Is this a transcription quality problem?

No, and that’s the useful distinction. Transcription quality is about whether the words are correct. Speaker identification is about whether the record knows who produced them. A transcript can be word-perfect and still attribute an entire council to a room. Paying for more accuracy won’t move it, because the missing ingredient is a way to tell voices apart, not better listening. A separate channel per speaker is one crude way to get that, and recognizing the voice itself is a better one.

Our clerk announces every name during roll call. Doesn’t that solve it?

Partly, and only while it’s happening. The names end up in the record because a person is reading them aloud in order, so recovering who voted means counting answers against that order. It works until two answers come at once, or somebody answers out of sequence, or the vote isn’t a roll call at all. Motions and voice votes usually aren’t. It’s a real mitigation and it isn’t a record.

Does this actually affect accessibility, or just recordkeeping?

Both, and they’re the same defect seen from two directions. Speaker identification is part of what makes a caption usable for a viewer who is deaf or hard of hearing, not a nice extra. It’s also what makes a transcript searchable and citable for anyone who wasn’t in the room. The same missing attribution shows up as a caption complaint and an archive complaint.

What does it take to fix?

A way for the system to tell one voice from another. The crude version is a microphone per seat, so each name gets its own channel; that works, and it costs equipment and a room rewire, and it still fails whenever two people share a mic. The better version is recognizing each speaker by voice, so the microphones you already have are enough. Either way, the first move is looking at your own transcript rather than pricing anything.

Should we be worried about how our names are spelled?

It’s worth ten minutes. Search two or three council surnames in your last transcript and count the variants. Inconsistent spelling is easy to miss, because a name that’s right most of the time looks correct at a glance, and it quietly breaks search for the resident trying to find what one specific official said.