Zumbo Download
Feature

Speaker labels, entirely on your Mac.

A transcript without labels is a wall of text. Knowing who said what is the difference between a record and a pile of sentences.

Why labels are the whole feature

Anybody can produce a transcript. The thing that makes one useful a week later is attribution: who committed to the deadline, who raised the objection, who said the number. Without that you have a document you still have to have been in the room to understand, which defeats the point of writing it down.

What actually happens

12:04

10:32Speaker 1We are holding the release until the migration lands.

10:33Speaker 2Then the changelog goes out Thursday, not Tuesday.

10:33Speaker 1Agreed. I will move the announcement.

Meeting mode running: a timestamp for each paragraph and a label for each speaker, written as the meeting goes.
  1. Turn Meeting mode on in the notch. Nothing is invited and nothing joins the call.
  2. Zumbo listens through your Mac's microphone, the same sound your own ears are getting.
  3. The transcript is written as the meeting goes, with a timestamp on each paragraph.
  4. Distinct voices are separated and each paragraph is labelled with the speaker it belongs to.
  5. Everything stays in Zumbo's history on the Mac, searchable.

The 220 MB add on

Almost all of Zumbo ships inside the download, which is why it is about 570 MB. Telling voices apart needs a second model, and it is an optional add on of about 220 MB that you download once. If you never turn on Meeting mode you never need it, which is why it is not in the main download making everyone pay for something most people will not use.

Once it is there it behaves like the rest: on your machine, no account, no upload, works with the network off.

Separating voices is not identifying people

Worth being precise, because the distinction matters. Zumbo works out that a meeting contained four distinct voices and keeps them apart. It does not know which one is Dana. You rename the labels afterwards, which takes a few seconds and is the only part of the process that requires you to have been there.

That is a deliberate limit rather than a missing feature. Building something that recognises specific people by voice means holding a voiceprint, which is exactly the kind of thing this app is designed not to have.

The other side of a call

Zumbo listens through the microphone, not to the audio your Mac is playing. On a video call, use the speakers instead of headphones: the microphone picks up the room, which now includes the other participants, and they land in the same transcript with their own labels.

Why this shape and not a service

The usual meeting assistant is a participant: invited, visible in the attendee list, recording to somebody else's systems. Zumbo does not attend, and there is nothing to hold, retain or subpoena, because nothing was received. Otter, Fireflies and Krisp promise not to misuse your meeting data. Zumbo never receives it.

Related: local meeting transcription for Mac, the Otter alternative that never leaves your Mac, and how Zumbo keeps your voice on your Mac.

Questions
How does it tell speakers apart?

By voice, on your Mac. The labelling model is an optional add on of about 220 MB that you download once; after that it runs locally like everything else.

Does it know people's names?

It separates voices, it does not identify them. You get one label per distinct speaker and you can rename them to the actual people afterwards.

Does a bot join the meeting?

No. There is nothing to invite and nothing appears in the attendee list. Zumbo listens through the microphone of the Mac you are already sitting at.

How does it hear the other side of a video call?

Use the speakers instead of headphones. The microphone picks up the room, including whatever the call is playing, and it all lands in the same transcript.

What if two people talk at once?

The same thing that happens to a human note taker: the overlap is messy. It recovers on the next clean stretch of speech, and the timestamps make it easy to find the part you need to check.

Where does the transcript live?

In Zumbo's history on your Mac, searchable, alongside everything else it has written. It is not uploaded and there is no account for it to be kept in.

Free for 3 days, everything unlocked. Then $15 once. macOS 14 or later, Apple silicon.