Record system audio on a Mac without a virtual driver

You can record the audio playing on your Mac by giving one app permission to listen to system audio, then pressing record. Since macOS 14.2, apps can tap into what other apps play without a virtual audio driver, so there is no kernel extension to install and nothing to route by hand. If you want the result as a transcript rather than a bare WAV file, an app that also listens to your microphone can save both sides and write the text out as it goes.

What counts as system audio on a Mac?

System audio is everything your Mac plays through its output: a call in Zoom or a browser, a video, a podcast, the other person on a FaceTime call. It is separate from your microphone, which is only your own voice and the room around you. macOS keeps the two apart on purpose, and for years it gave apps no clean way to record the output at all. That is why "record system audio" has always been a slightly awkward search. The obvious tools record your mic, not the sound coming out of your speakers.

Why was this always so hard?

For a long time the only route was a virtual audio driver. You installed a small piece of software that pretended to be a speaker, told macOS to send system sound into it, and then pointed a recorder at that fake device. It works, but trust me, even for someone who is comfortable in a terminal this was a nightmare to set up, and it always breaks with the next macOS update or resets itself when you plug in headphones. Apple's own screen recorder and QuickTime never captured system sound on their own, which is why every guide ended up recommending a driver plus a second app.

macOS 14.2 changed the starting point. Apple added Core Audio process taps, a supported way for an app to listen to what other apps are playing once you grant permission. No kernel extension, no rerouting, no fake speaker. An app asks for System Audio Recording permission the way it asks for the microphone, and from then on it can record the output directly.

How do you record it now?

You grant permission once and press record. Transcriber is a menu bar app that uses that tap: left-click the pill in the menu bar, or press ⌃⌥⌘R, and it starts capturing the system audio and your microphone together. The pill turns red and pulses while it runs, and a floating bar shows the transcript forming if you leave that on. Click again to stop and give the recording a name. The first launch asks for Microphone and System Audio Recording permission in System Settings, and the first recording downloads the transcription models, about a gigabyte, once.

Because it listens at the system level, it does not care which app made the sound. A meeting in Zoom, Teams or Google Meet, a WhatsApp or FaceTime call, a video in a browser, anything that reaches your speakers can be recorded. That is the same mechanism behind recording a meeting without a bot joining: the audio is already on your Mac, so you record it there instead of sending a notetaker into the call.

Do you get two separate channels?

Yes, and that is the part a plain system-audio recording misses. Your microphone and the system output stay as two channels, so your voice and the other side never land on the same track. Everything from the microphone is labelled as you, and everything from the system side is the other participants. After the recording stops, a diarization step runs over the system channel and splits it into Speaker 1, Speaker 2 and so on when it hears more than one remote voice. You get labels rather than real names, but you can type the names in during or after the call.

There is one wrinkle without headphones. Your microphone also picks up the sound from your speakers, so the far end can show up twice in the text. An echo gate uses the system channel as a reference and mutes the microphone frames that only contain the other side, and a text check drops the leftovers. Headphones are still the cleaner setup, and the gate is there for the calls where you forgot them.

What files do you end up with?

Each recording becomes a folder named by date and title. Inside is a readable Markdown transcript grouped by speaker with timestamps, and a line-by-line JSON file that is written and flushed to disk after every segment. That second file is the crash-safe copy: if the Mac quits mid-call, everything up to the last line is already saved, and you are offered the chance to finish the session on the next launch. Raw 16 kHz WAV files for the microphone and the system side, plus JSON and SRT exports, are optional if you want them.

All of it happens on the Mac. The only time it reaches the network is that one model download and a daily check for updates on GitHub, so the audio and the transcript stay on your disk. If you have used a cloud recorder before, that is the main change: nothing is uploaded, and there is no account.

When is a plain recorder the better call?

When you only want the sound. If the goal is a clean audio file of a track or a stream and you do not need the text or the speaker split, a screen recorder or an audio recorder paired with a virtual driver does that job and is often quicker to reach for. Recording both channels for a transcript makes sense when the recording is a conversation you will want to read back, search, or hand to an AI tool for a summary.

A couple of honest limits either way. This needs macOS 14.2 or newer, because that is where the tap arrived, and an Apple Silicon Mac is recommended for the transcription, so there is no Windows or phone version. And the transcript is a model's best effort, so names, product terms and strong accents are worth a read-through before you send anything on. How the app records and transcribes a call walks through the rest.