Blog · 2026-09-12

How to record both sides of a call on a Mac

You record a call on your Mac, play it back, and only your own voice is there. That is not a setting you missed. macOS does not hand the audio one app is playing to another app that asks for it. Here is why, the three ways around it, and why two separate tracks are worth more than one mixed file.

This is the practical version. If you want the Core Audio details with code, read recording the other side of a call on macOS instead.

Why you only get your microphone

In QuickTime, a new audio recording lets you pick an input device: the built-in microphone, a headset, an interface. The sound that Zoom or Chrome is playing is not an input device. It belongs to that app’s process, and macOS treats it as that app’s business.

This is deliberate. If any program could quietly capture what other programs play, every video call and every voice message on the machine would be readable by anything you installed. The restriction is the reason the workarounds below exist.

Option 1: a virtual audio driver

The long-standing approach uses a virtual device such as BlackHole, Soundflower or Loopback. You install a device that is both an output and an input, send the system output to it, and record it back as if it were a microphone.

It works. Three things make it tiresome:

• It is a system extension. On a managed work Mac, installing one is often blocked.
• You have to redirect your output device. Send audio only to the virtual device and you stop hearing the call yourself, so you also need a multi-output or aggregate device.
• You get one mixed file. Your voice and the other side land in the same track.

Option 2: screen recording with system audio

Recent macOS versions can capture system audio as part of a screen recording. For a tutorial video that is exactly right. For meeting notes it is awkward: you produce a video file you do not want, and the audio arrives mixed again.

Option 3: Core Audio process taps

Since macOS 14.2 there is a supported mechanism for this. An app describes which processes it wants to capture, the system asks you for permission (“System Audio Recording”), and after that the app receives the audio buffers.

What that buys you:

• no driver and no system extension to install,
• no output device to redirect, so the call keeps playing normally,
• a visible permission in System Settings that you can revoke at any time,
• and the important one: the other side arrives as its own track, separate from your microphone.

There is one quirk worth knowing if you are building this yourself. A tap on its own produces nothing; it has to be wrapped in an aggregate device whose main sub-device is your current output device, which supplies the clock that drift compensation keeps the tap aligned to.

Why two tracks beat one mixed file

Once you want notes and not just a recording, separation is the part that pays. From a mixed track, software has to guess who is speaking. Diarization is good and occasionally wrong, particularly with similar voices, people talking over each other, and poor connections.

With your microphone on one channel and the call on another, attribution stops being a guess and becomes a property of the recording. For notes someone else reads later, the gap between “the client probably said this” and “the client said this” is not small.

Three practical notes

• Wear headphones. Play the call through speakers and your microphone picks up the other side as well. The separation is then only clean on paper.
• Watch the disk. Uncompressed, an hour of conversation runs to a gigabyte or two. AAC is plenty for speech. We switched and now use about one twenty-sixth of the space; on a real twelve-minute meeting the transcript matched the uncompressed one 99.6% character for character, and the differences were punctuation.
• Say that you are recording. This method produces no notification for the other side. That is exactly why it has to come from you. See does Zoom tell people you are recording?

The short version

For occasional captures, a virtual audio driver is fine. If you are recording meetings regularly and want notes out of them, process taps are the better path: nothing to install, a permission you can take back, and both sides of the call kept apart. That is how Cachalot records, and because transcription and the notes also run on the Mac, the recording never leaves the machine it was made on.