Private speaker-aware transcription

Know who said what.
Entirely on your Mac.

Meetings, interviews and lectures become searchable transcripts that identify not only what was said, but who said it. Every model is already inside the app.

Download on theMac App Store

USD 39.99One time · No subscription

macOS 14 or later · Apple silicon + Intel · No account

Quarterly review.m4a

42:18 duration88% speech coverage

AMAmara · 18:42

Let’s start with the decision we need to make today.

JLJon · 13:08

The customer interviews point in one clear direction.

MKMika · 10:28

I’ll capture the next steps and send them this afternoon.

Amara

Jon

Mika

Raw transcription gives you words.

MinuteMark gives the conversation structure.

Voices become people

MinuteMark locates speech, transcribes it, matches each passage to a voice and groups those passages into speakers. Every speaker gets a colour and a label you can rename.

Rename once

Change “Speaker 2” to a real name and it flows through the transcript and every export. Speaker identity stays consistent from review to handoff.

You set the boundaries

Tell MinuteMark the head count, or let it work the number out. A sensitivity control decides how readily similar voices are treated as one person.

See who had the floor

Talk time is shown for each speaker in minutes and as a share of the conversation. Speech coverage shows whether a recording was mostly talk or mostly silence.

A transcript made to read

Consecutive fragments from the same person are merged into natural turns, so one paragraph is not split into a dozen rows. Search finds any word across the transcript.

Take it anywhere

Export Markdown with a talk-time table, plain text, WebVTT with speaker voice spans, CSV or structured JSON.

Offline by architecture

Nothing to download.
Nowhere to upload.

Every speech and speaker model ships inside MinuteMark. There is no account to create and no network connection involved in processing.

Models bundled in the app
Recordings stay on your Mac
No account, analytics or telemetry

HONEST OUTPUT

It would rather say “unknown” than be confidently wrong.

A recording with no speech is reported as such. A stretch that cannot be attributed to a voice is dropped instead of being assigned to the wrong speaker.

Five useful formats

From conversation
to working document.

.MDMarkdown + talk-time table

.TXTPlain text

.VTTSpeaker voice spans

.CSVRows for analysis

.JSONStructured data

MinuteMark for macOS

The room spoke.
Keep track of who.

One purchase. No subscription. No account.