Diphony — Mac meeting recorder with real speaker labels

How it works

Two tracks that never touch each other

of turns landed on the wrong speaker — eight test scenarios and four real recordings

names in the transcript when a leading single-track app transcribed the same call

45s

to turn an hour-long meeting into a transcript, on an M1 Pro

Your voice comes in through the microphone. Their voice comes out of your speakers. Diphony keeps those two on separate files from the first second — so knowing who said what is a fact about the recording, not a guess about the sound.

Your microphone and your Mac's audio stay on two separate tracks through echo removal and on-device transcription, producing a transcript labelled by speaker

Our own measurements: eight scripted test scenarios, plus four real recordings made on 1 September 2026 with a person speaking into the microphone while the other side came out of the speakers. The comparison is a leading single-track transcriber reading that same audio mixed down to one track. Speed measured on an M1 Pro — 29–34 s of transcription plus 11 s of echo removal per hour of meeting. Small samples, and we will say so until they are bigger.

The problem

One recording can't tell two people apart

Most tools capture the call as a single mixed track, then work backwards to guess who said what from the sound of the voices. That holds until someone speaks quietly, until two voices are similar, until people talk over each other — which is most meetings.

One mixed track

Speaker labels are a guess, and guesses drift mid-call
Your own voice, bleeding out of the speakers, gets counted twice
A quiet participant can vanish from the transcript entirely

Diphony's two tracks

Your side and their side are separate files from the first second
Attribution is a fact about which file the audio came from
Speaker echo is removed before a single word is transcribed

What you get

A transcript you can actually work with

Click a line, hear that line

Open a session and click any sentence to play exactly that moment. Press play at the bottom for the whole call — the highlight follows the audio as it moves, so you never lose your place.

⌘F searches inside the open session

Transcript lines labelled ME and SARAH CHEN in two colours, with an EchoGate badge showing 19 dB of echo removed

Folders that behave like Finder

Drag sessions into folders, select several and move them at once, star what matters, and undo any of it with a keystroke. Name the other speaker once and it sticks — on every line and in every export.

⌘⇧N new folder · ⌘Z undo a move

The Clients folder selected in the sidebar, with the two sessions filed inside it listed to the right

Privacy

It stays on this Mac — by construction, not by promise

Recording and transcription both run locally. Every session is an ordinary folder on your disk — two audio files and the transcript — in a location you choose. There is no server to upload to, no account to create, and nothing for us to look at.

the only two addresses the app contacts: the App Store, and huggingface.co to fetch the speech model once

analytics SDKs, crash reporters and advertising networks in the app

100%

of transcription runs offline after that first model download

Details in the Privacy Policy.

Pricing

Free is genuinely usable

Every feature is unlocked on the free plan. The only limit is how much you record — and sessions you already made stay yours forever.

Free

no account, no card

2 sessions a day
Up to 60 minutes each
Every feature unlocked
Export and search included

3-DAY FREE TRIALPro · Monthly

$2.99/mo

try it free for 3 days

Unlimited sessions, any length
Everything in Free
Cancel any time in the App Store

Pro · Lifetime

$29.99 once

pay once, yours forever

Unlimited sessions, any length
Everything in Free
No subscription, nothing renews

Both Pro options unlock exactly the same thing — pick whichever way of paying you prefer.

Prices in USD; the App Store charges in your local currency. Subscriptions renew automatically unless cancelled at least 24 hours before the period ends — see the Terms of Use.

The details

Everything in one table

Requires: macOS 15 or later · Apple silicon and Intel
Records: Your microphone and your Mac's system audio, as two separate files
Attribution: By which track the audio came from — not by matching voices
Transcription: On this Mac · English
Network use: The App Store, and huggingface.co once for the speech model
A session is: A folder holding the two audio files and the transcript, where you choose
Export: Markdown or plain text
Account: None — nothing to sign up for
Analytics: None · no SDK, no crash reporter, no ad network
Free plan: 2 sessions a day, up to 60 minutes each
Pro: $2.99/month with a 3-day trial, or $29.99 once
Sold through: The Mac App Store

Questions

Before you download

Does it work with Zoom, Teams, Google Meet?

Yes. Diphony records whatever plays through your Mac's audio, so it does not care which app the call is in. It never joins or dials a call itself — it listens to your own machine.

Do I need to tell people I'm recording?

Often yes. Recording a conversation may require the consent of the other participants, and the rules differ by country and by state. Diphony gives you the tool; obtaining permission is your responsibility.

What happens when a free session hits 60 minutes?

It stops on its own, and the part already recorded is kept and transcribed — you never lose what was captured. Pro removes both the daily count and the length limit.

Where are my recordings kept?

In a folder you choose, shown in Settings. Each session is a plain folder holding the two audio files and the transcript. Delete it in the app or in Finder and it is gone.

Does transcription need internet?

Only the first time, to download the speech model. After that transcription runs entirely offline on your Mac.

Which languages does it transcribe?

English. The on-device model is English-only for now.

Every line, the
right speaker