A transcript that separates voices correctly but still calls them "Speaker 1" and "Speaker 2" has done half the job. The other half is a name — and getting it wrong, or leaving it generic, is the kind of small thing that makes a transcript feel unfinished even when the substance of it is exactly right. Fixing it is quick, and it's worth understanding why it only has to be done once per person rather than once per meeting.
- 00:01:04Speaker 2I can take the follow-up with legal this week.
- 00:01:04DanaI can take the follow-up with legal this week.
Why a label starts generic in the first place
Speaker separation works by comparing how different voices sound against each other during a recording, which is enough to tell two people apart without knowing either of their names. The app has no directory to check a voice against on its own — nobody typed in who's attending before the meeting starts — so the first time a new voice shows up, it gets a placeholder label rather than a guess that might be wrong.
Step 1 — find the line where the name got mislabelled
Open the transcript for the meeting in question and scroll to a point where the speaker is clearly identifiable — someone introducing themselves, being addressed by name, or a section where context makes it obvious who's talking. That's the anchor point for the correction, not a random line picked at random from partway through the call.
Step 2 — rename the label, not just that one line
Click the speaker label and enter the correct name. This updates every instance of that label across the entire transcript, not just the line that was clicked — a placeholder like "Speaker 2" and every line attributed to it becomes the corrected name in one action, because the label and the voice behind it are the same thing throughout the recording.
Step 3 — check the minutes reflect the change too
Because the generated minutes are built from the transcript, a name corrected there is what shows up wherever that person is referenced in the summary as well — an action item that read "Speaker 2 to follow up with legal" becomes "Dana to follow up with legal" once the underlying transcript carries the right name.
Step 4 — confirm the correction carries into the next recording
The first time a voice is matched to a name, that match is what the app checks future recordings against — a voice print built from what was already captured, not a fresh guess starting from scratch each time. The next meeting with the same person in it should open already labelled correctly, which is the actual payoff of doing the correction properly the first time rather than skipping it because "it's just this one transcript."
What happens when two different people get merged into one label
Occasionally two voices that sound close enough — over a laptop microphone, on a bad connection — get treated as one speaker rather than two. The fix is the same mechanism as a plain rename, applied per line rather than to the whole label: correct each mislabelled line to the right person as you notice it, and the distinction improves in later recordings as more of each voice gets captured cleanly and separately.
What happens when one person's voice gets split into two labels
The opposite error is less common but does happen — background noise, a cough, or a change in the microphone partway through a call occasionally reads as a second voice. Renaming both labels to the same person's name merges them back together in the transcript, and it's worth doing before the minutes are generated, since an action item attributed to a name that doesn't exist elsewhere in the transcript is a small but avoidable inconsistency.
Why this matters more than it seems to
A transcript with the right names in it isn't just more pleasant to read — it's more useful to search later. A search for "what did Dana say about the renewal" only works if "Dana" actually appears in the text rather than "Speaker 2." The habit of correcting names as they come up, rather than leaving a transcript in placeholder form because the substance is already right, is what makes an archive of meetings searchable by the details that actually matter months later.
Doing this the first time a new person joins a recurring meeting
The moment worth doing this properly is the first meeting with someone new in it — a new hire on a recurring team call, or a client's colleague joining for the first time. Correcting the label in that first recording means every later meeting with the same person opens correctly labelled from the start, rather than accumulating a backlog of transcripts that all need the same correction applied separately.
What this looks like for someone running a lot of first-time calls
A recruiter running back-to-back screening calls meets a new voice in nearly every recording, which means there's rarely a "next time" this correction pays off the way it does for a recurring meeting — each transcript needs its own candidate's name entered once, for that one conversation. That's still worth doing rather than skipping: a transcript that says "the candidate mentioned they're available from March" is less useful later than one that says the candidate's actual name, especially once several weeks of interviews have piled up and the names are what separates them from each other.
Correcting a name after the meeting, days later
There's no window in which this has to happen. A transcript from three weeks ago can be corrected the same way as one from this morning — renaming a label is a text edit against a file that already exists, not a live step that has to happen while the recording is fresh. The only cost to waiting is that the placeholder label sits in the archive, unsearchable by the real name, until the correction is made.
The difference between fixing a name and fixing a misheard word
This page is about a speaker's identity, not the accuracy of what was transcribed word for word. A name mistyped into the transcript because it was misheard — "Dana" transcribed as "Donna" in the actual spoken text — is a different, smaller correction than relabelling who a whole block of dialogue belongs to, and it's worth telling the two apart: one is who said it, the other is what they said.
Why this doesn't require re-transcribing anything
Renaming a speaker label is a correction to how a transcript is displayed and attributed, not a request to run the recording through transcription again. The underlying transcription work — turning audio into text, segment by segment, while the meeting happened — is already done and stays exactly as it was; only the label changes, and it changes for every line at once rather than requiring the transcript to be regenerated.
Keeping this from happening again in the same meeting
If a meeting has several unfamiliar voices in it — a panel, an interview with more than one interviewer — correcting names early in the recording, rather than waiting until the end, means later automatic matches within that same call have a better chance of landing on the right label the first time, since more of each voice has already been distinguished from the others by the point new segments are being attributed.
Why a wrong name is worse than a generic placeholder
A label that says "Speaker 2" is honestly uncertain — it doesn't claim to know who's talking. A label that says the wrong name is confidently wrong, which is a different and more damaging kind of error: it can end up in minutes sent to other people, attributing a decision or a commitment to someone who never made it. If there's any doubt about who's actually speaking, leaving the placeholder in place until it's certain is safer than guessing and correcting it to a name that also turns out to be wrong.
When someone's name changes — a title, a preferred name, a correction
People's names aren't static. A colleague changes how they'd like to be addressed, someone's title changes and gets referenced differently in later meetings, or an earlier correction turns out to have used a nickname rather than the name that should show up in formal minutes. Renaming a label again later works the same way as the first correction — it updates every instance of that speaker across the transcript being edited, and going forward, the newer name is what gets used.
Correcting a name across a long-running series of meetings
A weekly meeting that's been running for six months accumulates a lot of transcripts, and a name corrected today only affects recordings from today onward — it doesn't retroactively rewrite transcripts that already exist from before the correction was made. If a name was wrong from the very first meeting in a series, older transcripts in that series keep the original placeholder or incorrect name unless each one is corrected individually; only new recordings benefit automatically once the voice is matched.
Who ends up doing this the most
Anyone whose meetings are full of new voices notices this step the most. A recruiter running back-to-back interviews corrects a name nearly every single call, since there's rarely a recurring voice to already be matched. Someone on a stable team with the same six people in every meeting, by contrast, does this correction once per person, total, and then benefits from it silently for as long as that group keeps meeting the same way.
A mislabeled name discovered well after the fact
If a transcript from months ago turns out to have had a name wrong the whole time — noticed only when it's read again for some other reason — the correction still works the same way it would on a fresh recording. There's no expiration on fixing it; the only cost of waiting is that anything searched or referenced from that transcript in the meantime carried the wrong name until the correction was made.
Why this is a manual step rather than an automatic guess
The app deliberately doesn't try to guess a name from context — from what someone says about themselves, or from a calendar invite it has no access to in the first place, since there's no calendar integration at all. Matching a voice to a name only happens once a person has confirmed it, which means the system never confidently attaches the wrong name to a voice without a human having said so first. That's a slower first step than an automatic guess would be, and it's also a safer one: a wrong guess presented confidently is a worse outcome than a placeholder waiting for a human to resolve it.
A worked example, start to finish
A recurring client call has three regular attendees plus, this week, a new person from the client's side who's never joined before. The transcript labels the three familiar voices correctly from the first line, because they were matched in earlier calls, and labels the new voice "Speaker 4." Partway through, that person is introduced by name. Clicking their label and entering the name updates every line attributed to them in this transcript, and the minutes generated afterward show their name rather than the placeholder in any action item assigned to them. The next time that same person joins a call, their voice is already recognized from the start — the one correction made this week is the only one that meeting ever needs.
What it costs, what it runs on
Renaming a speaker, and having that correction carry forward automatically, is built into the same $19 one-time license as recording and transcription — no separate step to unlock, no subscription tier that gates it. It works on macOS 13 Ventura or later, on up to three of your own Macs, with a 30-day trial that includes this from the first recording rather than holding it back for later.