How to record your microphone and your Mac’s system audio at the same time

You need two devices in Audio MIDI Setup, not one: a Multi-Output Device so the sound your Mac plays goes to your headphones and to a virtual audio driver at the same time, and an Aggregate Device so a recorder can see that driver and your microphone as one input. Miss the first and you record the call but hear nothing; miss the second and you record one of the two sources and find out afterwards.

Or you skip the whole arrangement, which is the last section.

The signal path, written out

Read it in two hops.

Playback hop. Every app’s output → Multi-Output Device → your headphones + BlackHole (or Loopback, or whatever virtual device you installed).

Capture hop. BlackHole + your microphone → Aggregate Device → your recording app.

Apple describes each half in one sentence. For the Multi-Output Device: audio sent to the primary device “also plays through any other device in the stack.” For the aggregate: “When you create an Aggregate Device, multiple devices appear as a single device in an app.”

Step 1 — build the Multi-Output Device

1. Open Audio MIDI Setup (it’s in /Applications/Utilities, or ⌘Space it).
2. Click + at the bottom of the sidebar → Create Multi-Output Device.
3. Tick Use for your real output first — the built-in speakers or your headphones — and then for BlackHole. Order matters. BlackHole’s own wiki is blunt about why: “Due to issues with macOS the Built-in Output (sometimes called MacBook Pro Speakers or similar) or another 2 channel audio device must be enabled and listed as the top device / primary device / clock device in the Multi-Output.” If BlackHole ended up on top, untick and retick to reorder.
4. Enable Drift Correction for every device except the clock source, which is the top one. That is the wiki’s recommendation and it matches Apple’s: drift correction is “also known as resampling, to compensate for drift in the data between devices.”
5. Set your Mac’s sound output to this new device.

You will now hear audio normally and BlackHole will be receiving a copy. If you only wanted the call’s audio, you can stop here, point a recorder at BlackHole as its input, and you’re done — that’s the route in recording system audio on a Mac.

Step 2 — build the Aggregate Device

Your microphone is still not in the picture. BlackHole carries output, and your recorder can only choose one input, so the two have to be presented as a single device.

1. In Audio MIDI Setup, click + → Create Aggregate Device.
2. Tick Use for BlackHole and for your microphone. The order you tick decides the channel order your recorder sees.
3. Set the Clock Source to your microphone — the physical device with the real crystal, not the virtual one.
4. Enable Drift Correction on the other member.
5. Check that both are at the same sample rate. Apple states this as a requirement rather than a suggestion: “All devices in the aggregate device need to have the same sample rate.”
6. In your recorder, choose the Aggregate Device as the input.

That last check is where a lot of chipmunk recordings come from. BlackHole’s rate is whatever Audio MIDI Setup says it is, and a 2020 issue on the project reports it “doesn’t seem to set the sample rate based on the actual stream as requested by an application.” Mismatch the two members and the arithmetic of playback goes wrong — see why a Mac recording plays at the wrong speed.

The part the tutorials skip

You have an aggregate device with, say, five channels. Your recorder wants two.

An aggregate does not mix its members, and almost nothing written about this setup says so. BlackHole’s maintainer, answering a user trying exactly this combination of system audio plus a Yeti microphone: “You need to record all channels as an aggregate does not mix them,” and “Garage band only is able to record 2 channels. Your aggregate has 5.”

So a naive setup fails in one of two shapes, and both look like a bug:

• Only one source is in the file. A two-channel recorder grabbed channels 1–2, which belonged to whichever device you ticked first. Your voice or theirs, never both.
• One side is silent, or sits entirely in the left channel. The channels arrived, got interpreted as ordinary stereo, and the layout was never what your recorder assumed.

The fix is a recorder that takes every channel of the aggregate and mixes down deliberately — a DAW, Audacity with the right input configuration, Loopback’s own patchbay, OBS with per-source tracks. QuickTime is not that recorder. If your recording came out silent instead, the causes are enumerable and different: diagnosing a silent Mac recording.

What this arrangement costs you, honestly

The routing works. Living with it is the problem.

You lose keyboard volume control. Not a bug you can fix: “Unfortunately macOS does not support changing the volume of a Multi-Output device but you can set the volume of individual devices in Audio MIDI Setup.” Mid-call, that means opening a utility to turn someone down.

Two clocks drift. Drift correction resamples to compensate, which is why Apple ships the checkbox, and on a long recording the compensation is the difference between usable and slowly-sliding audio.

There is a ritual. Switch output before the call, switch back after. Forget the second half and your music plays through a device that no longer has a virtual sink attached, or your next Zoom call hears nothing.

The result is one track. Even done perfectly and mixed down properly, you end up with a single waveform containing several people. That matters less if you want the audio and more if you want a transcript, which is the next section.

Why two tracks beat one mixed track

Say you want notes rather than a recording. A mixed track gives a transcriber one stream of overlapping voices, and everything about attribution becomes inference: who spoke, where one person stops and the next begins, which lines were yours.

Two independent streams change the shape of the problem. Your microphone is, by definition, you — so those lines can be attributed with certainty rather than clustered and guessed at. The call’s audio is everyone else, and speaker separation runs on that track alone, where the voices are the ones that need separating.

This is how Narada records. It holds macOS’s System Audio Recording permission and reads the output stream where it already is, so there is no Multi-Output Device, no Aggregate Device, no driver to install and nothing to set back afterwards. The microphone and the system output are written as separate .m4a files and transcribed separately, on the Mac, using Parakeet TDT v3 through Core ML. Speaker labels on the call’s side are Speaker 1, Speaker 2 and so on — the model separates voices, it does not know names, and the app doesn’t pretend to. Your own lines carry your name only because your microphone was its own file.

Two limits worth knowing. Changing audio devices mid-recording — pulling out AirPods, plugging in a dock — ends one part and starts another, with a two-to-five-second gap at the seam; the parts present as one meeting, but the seam is real. And macOS audio can fail underneath any app. Narada checks each finished recording against what the capture actually submitted and marks it Incomplete, with a reason, when audio went missing. It cannot get that audio back. Nothing can. The useful part is being told, rather than finding out three weeks later that a 90-minute meeting is six minutes long.

The recorder half is free — capture, transcription, speaker separation, plain files on disk. Summaries and follow-ups are the paid part, or free on-device on an Apple Silicon Mac after a one-time model download of about 3 GB or 7 GB depending on your RAM. It needs macOS 15.6 or later.

Which route to take

One thing no software settles for you: if the audio you are capturing contains another person, tell them. Rules differ by country and by state, and some places require everyone’s consent rather than only yours. Narada keeps a reminder on screen for the length of the recording, and that is the extent of what an app can do about it — it doesn’t notify anyone on your behalf and it can’t make a recording lawful.

Sources

• Apple Support, Combine audio devices into a single aggregate device in Audio MIDI Setup on Mac — the definition and the creation steps.
• Apple Support, Set aggregate device settings in Audio MIDI Setup on Mac — drift correction “also known as resampling"; the same-sample-rate requirement.
• Apple Support, Play audio through multiple devices at once in Audio MIDI Setup on Mac — the primary device and the stack.
• Apple Support, Create an Aggregate Device to combine multiple audio devices — “multiple devices appear as a single device in an app”.
• Apple Support, Record audio in QuickTime Player on Mac — the single input control, “Choose a microphone (if you have more than one available)" — https://support.apple.com/guide/quicktime-player/record-audio-qtp3a7cbdc0d/mac
• Apple Support, Take a screenshot or screen recording on Mac — “Below Microphone, select the microphone you want to use”, with no system-audio option — https://support.apple.com/guide/mac-help/take-a-screenshot-or-screen-recording-mh26782/mac
• BlackHole wiki, Multi Output Device — primary-device ordering, drift-correction advice, and the volume limitation.
• ExistentialAudio/BlackHole, Discussion #716 — https://github.com/ExistentialAudio/BlackHole/discussions/716 — “an aggregate does not mix them”; “You need to record all channels”.
• ExistentialAudio/BlackHole, Issue #41, Sample rate not set by application? (January 2020, closed) — https://github.com/ExistentialAudio/BlackHole/issues/41 — BlackHole’s sample rate not following the source stream.
• Existential Audio, BlackHole product page — free, donation-based, “Compatible with macOS Mavericks (10.10) and newer”, Intel and Apple Silicon.

Every one of these was fetched and checked on 4 August 2026.