Speaker profiles
Abschrift separates speakers within every meeting. Turn speaker recognition on, and it also remembers voices across meetings. This page explains how that works and how to keep your speaker list tidy.
It's biometric data — so it's opt-in
Recognizing a person by their voice makes the stored fingerprint biometric data, the most protected category under modern privacy law. Abschrift therefore ships with cross-meeting recognition off. You can turn it on during onboarding, or later in Settings → Speaker recognition. With it off, every meeting is still fully separated into speakers, labeled “Speaker 1” and “Speaker 2” within that meeting. No voice is stored or matched across meetings. If you record colleagues or clients regularly with recognition on, tell them in the same announcement as the recording itself. See Recording & the law.
How voice memory works
During processing, each remote speaker's voice gets a compact acoustic fingerprint — a
set of numbers, not a recording. Abschrift compares new voices against your stored
profiles. A match fills in the name automatically. A voice that matches no profile becomes
a new profile named “Speaker N”. All of this happens on your Mac: fingerprints live in
~/Abschrift/speakers.sqlite and never leave it.
Your own voice is a special case. The microphone track is always attributed to You.
Renaming
Open Speakers from the sidebar footer and type over a name. Renames propagate into all past meetings — transcript labels, the participants list, and search. History always reflects what you know now. Inside notes, only auto-assigned “Speaker N” names are rewritten. Sentences that mention real names stay untouched.
There is a shortcut where you'll actually meet a new voice. In a meeting's header, an unnamed participant like Speaker 4 is underlined. Click it and type the name right there. It's the same profile rename, so it updates your whole history too.
Fixing a wrongly assigned segment
Recognition is good, but no speaker separation is perfect. Occasionally a sentence lands on the wrong person. Fix it where you see it: in the transcript, click the speaker label on that segment and pick who really said it. You can pick another participant, any stored profile, or a name you type. Tick “All … segments in this meeting” to relabel everything that voice was credited with in this meeting.
One distinction is worth knowing: reassigning edits the label in this meeting's transcript only. It deliberately does not touch any voice fingerprint. So when your edit looks identity-shaped, Abschrift offers the profile operation instead. Relabel all of an unnamed speaker's segments onto someone it already knows, and it offers a merge. Relabel them onto a name it has never heard of, and it offers a profile rename — the operation that updates past and future meetings alike. Decline, and the old profile keeps that voice — it would greet the person by the wrong name in your next meeting.
Listen before you decide
With keep recordings turned on, every timestamp in the transcript is a play button. You choose this during onboarding, or later under Settings → Storage & retention → “Keep recordings after processing”. Click a timestamp to hear exactly that segment, stopping where the next one starts. That's the fastest way to check who a voice belongs to before renaming or reassigning.
Meetings with playable audio say recording kept in their header. Kept recordings follow your retention schedule — once deleted, the play buttons disappear.
Merging duplicates
Occasionally one person ends up as two profiles — after a bad headset day, or a very short first appearance. Use Merge into… on the duplicate. The two voice fingerprints are combined, and the duplicate disappears. Its past meetings are relabeled with the surviving name. Future meetings recognize the person more reliably, because you gave Abschrift more voice evidence to work with.
Merging can't be undone. The two fingerprints are averaged into one, and the original pair no longer exists afterwards. Before anything changes, Abschrift asks you to confirm and tells you how many meetings will be relabeled. If you merge two people who are not the same by mistake, the combined profile will match both voices poorly. The way out is to delete it and let both people re-appear as new speakers.
Deleting
Deleting a profile removes the stored voice fingerprint. Past transcripts keep their labels. The person is treated as unknown next time and re-appears as “Speaker N”. That makes delete a safe reset button — including for a merge you regret.
Tips for best results
- Name speakers early — every named meeting strengthens recognition.
- Wrong label on a segment → reassign it in the transcript. Wrong identity (one person, two profiles) → merge the profiles. Wrong name on a profile → rename it. The app nudges you to the right one.
- If you want to verify voices by ear before renaming or reassigning, keep audio on.
- The same profile tools exist headlessly:
abschrift speakers list|rename|merge|delete.