Naming speakers

A transcript with speaker labels tells you there were four people. Naming them tells you which one promised to call the lender on Monday. It takes about thirty seconds, and it's the step that makes the transcript worth keeping.

Last updated 2026-08-31

Why the labels start as numbers

When you transcribe a file with speaker labels, LymeScribe separates the voices and tags each line with the one that produced it. What it cannot do is know a stranger's name from the sound of their voice — no software can. So the labels come out as Speaker 1, Speaker 2, and so on, and you supply the names.

What comes out

Speaker 1: ...so we should push
the closing to March.
Speaker 2: I don't think the lender
will go for that.
Speaker 3: Okay so who's calling them?
Speaker 2: I can do it Monday.

After thirty seconds

Mike: ...so we should push
the closing to March.
Phil: I don't think the lender
will go for that.
Susan: Okay so who's calling them?
Phil: I can do it Monday.

The Name speakers flow

Open the History window — the archive — and find the transcript. Every transcript-level action lives there. On its row, choose Name speakers (on Windows the button reads Resolve speakers…). A panel opens listing every speaker the transcript found.

For each speaker you get:

A play buttonPlays a short clip — up to about eight seconds — of that voice, taken from the original recording. This is how you tell who is who.

A sample lineOne quoted line that speaker said, so you can often identify them by content without playing anything.

A segment countHow many times they spoke. Useful for spotting the one-line participant versus the person who ran the meeting.

A name fieldType the real name. That's the whole interaction.

A merge menuFor when one person got split across two labels — see below.

Toggle Preview at the top to see the transcript re-rendered with the names you've typed so far, before you commit to anything.

The play button needs the audio. Clips are streamed from the original recording, so the app has to still have it. If audio wasn't kept for that file, the play button stays visible but disabled and says so — you can still name speakers from the sample lines and the transcript itself. Whether audio is kept is a retention setting; see History & the archive.

When one person becomes two speakers

Sometimes a single person gets split across two labels — they moved away from the microphone, or changed how they were speaking partway through. Use the merge menu on the duplicate row and choose the speaker to fold it into. Merged rows show the fold visibly and carry an Undo button, so it's a safe thing to try.

If this happens a lot in your recordings, Tuning speaker labels covers what to change about the recording itself.

Saving, skipping, resuming

Two buttons finish the job:

Copy Resolved puts the renamed transcript straight on your clipboard — the fast path when you're about to paste it into an email or an AI.
Save Resolved writes it out as a new file alongside the original, with _resolved appended to the name, and syncs the names back into your history.

Nothing here is mandatory. Close the panel without doing anything and the original transcript is untouched. But if you typed at least one name or made a merge, those edits are written back to history automatically when you close it — you won't lose work by closing the wrong thing.

And it's resumable. Reopen a transcript you partly named and the names you already assigned come back, with the action now reading Resume naming speakers. You can name two people today and the other three next week.

The names never leave your machine. Speaker names are yours, stored with the transcript on your own computer. They aren't sent to a server, and nothing about this step involves the network.

The corrections dictionary

Different problem, same family: a word that comes out wrong every single time. A product name, a colleague's surname, a piece of industry jargon the speech model has never met. Naming speakers fixes one transcript; corrections fix the word permanently, in everything you transcribe from now on.

Open Settings ▸ Corrections. The model is a group rather than a find-and-replace pair: one target term — the spelling you actually want — and as many misheard variations as you need mapping onto it.

Target termThe correct spelling, e.g. LymeStack.

VariationsEverything that should become it, e.g. Lime Stack, lyme stack, Limestack. Add as many as you discover.

Case sensitiveOff by default. Turn it on when the difference between two capitalizations matters.

Whole wordMatch only the standalone word, so a correction doesn't fire inside a longer word that happens to contain it.

Create one with the target box, the arrow, and the source box, then Add Group. Add further variations to an existing group at any time, remove one variation, or delete the whole group.

When corrections are applied

Always after transcription, and always on your own machine. The speech model produces its best guess, and then LymeScribe rewrites the terms you've taught it — which means corrections apply to dictation and to batch files alike, and adding a new correction never requires re-running anything on a server.

When you're connected to a LymeScribe Server, correction groups sync with it so a whole office can share one vocabulary. The Corrections pane shows when it last synced and has a Refresh button, and it works offline — an unreachable server shows an offline badge rather than blocking your corrections from applying.

Build it as you go. Don't try to write your whole vocabulary up front. Add a group the moment a transcript gets a word wrong for the second time; a few weeks of that and the app rarely misses anything that matters to your work.