Why the transcript has two labels and not names
A meeting transcript in Dictera has two labels: you, and them. Everyone who is not you shares the second one. Names appear in the notes only where you listed the people and the meeting itself makes clear who said what. That is a decision, and it comes from what the audio can and cannot tell anyone.
Where the two labels come from
A meeting reaches your Mac as two streams. Your microphone hears you. The call's audio, the sound your Mac plays through your speakers or AirPods, carries everyone else. Dictera transcribes each stream separately and merges them into one transcript, so the label on a line is simply which stream it arrived on.
That makes the two labels certain in a way a name never is. A line on the Them side was not said by your microphone's owner. Nothing had to be inferred to put it there.
The certainty has one edge worth knowing. "You" means everything your microphone heard. In a conversation across a table, the other person is in the room with your microphone, and their words land on your side. The assistant is told this: when you pick a You line to answer, it considers that the line may be the other person speaking, not you.
Why not tell the voices apart?
By the time the call reaches your Mac, the other side is already one mixed signal. Separating it into people means modelling each voice: what makes Anna sound like Anna, kept long enough to recognise her again. That is a voiceprint, built from people who never installed Dictera and never agreed to it. We chose not to build one, of anyone.
There is a quieter reason too. Voice separation guesses, and its guesses look exactly like facts. A transcript that says "Anna:" in front of a line Tom said is worse than one that says "Them:", because nobody rereads a name to check it. Two labels are less informative and never wrong about which side a line came from.
And nothing is kept to guess from later. The audio is transcribed as it arrives and never written to disk, so there is no recording to run a second, slower pass over after the call.
So where do names come from?
From you, or from the invite. A meeting's Setup tab, on the Meetings page in the notch, has a field for the other participants, Who else is in the meeting. Type the standup's regulars once, press Save as my defaults, and new meetings start from that list. If you have connected your calendar, the names on the invitation fill it in: organiser first, then the people who accepted, then the rest, with anyone who declined left out.
The list tells the notes who was there. It does not tell them who spoke which line, and the model is instructed accordingly. It uses a name only when the transcript makes clear who is speaking or who is being addressed: somebody says "Anna, can you take this?", or a person introduces themselves. An action item whose owner the meeting never established is written Unassigned, never handed to the likeliest-looking name.
What happens when the invite is incomplete
Calendars often know an invitee only by email address. Those people have no name to put in the field, and Dictera does not store addresses at all. So the list can be short: eighteen names for twenty people, shown as a small "18 of 20" badge beside the field.
A short list is more dangerous than an empty one. Given eighteen names and no warning, a model reads "not listed" as "not present" and gives every commitment to someone on the list. So when the list is partial, the notes are told so outright: more people were present than are named, and no speaker should be assumed to be one of the names just because those are the only names available.
The same caution runs through the exported notes. The header says who was invited, never who attended, because the calendar only knows who was asked. It gives no count of how many people spoke either. The transcript cannot tell two voices on the Them side apart, so any number there would be made up.
Where the names go
Names typed into the field are yours to send wherever your notes are written. Names Dictera read from your calendar have a switch of their own, Send attendee names to remote engines, under Privacy in Settings ▸ Calendar. It starts off, so if you point Meeting Notes at a remote engine, those names are removed before the request leaves your Mac. Switch it on, and they go with the request. The security page lists this with everything else that can leave the machine.
The cost, honestly
You cannot scroll a transcript and see who said each line. In a one-to-one call that costs nothing, since Them is one person. In a meeting of eight, it means the transcript tells you what was said and roughly when, and the notes tell you who owns what only where the meeting said it out loud.
We think that is the right trade, for the same reason the notes are not written live: a document you can trust beats one that looks more complete. What Meeting Notes does with the names you give it is on the Meeting Notes page.
