How to Record Internal Audio on Mac (2026)

Here's the thing about macOS audio: it is deliberately locked down. Your apps make sound, but capturing that sound — without an extra driver, a paid utility, or a clever framework — has been a hassle for years. The good news is that there are now several reliable paths. The better news is that one of them keeps your voice and the other side's voice on separate tracks, which is the only sane way to capture a meeting you actually want to read later. This guide walks through every method I know that works on macOS 14 and 15, on Apple Silicon. I run MetaWhisp daily and tested each path on my own M-series Air while building it. The version you have on your desk and the version of macOS you have on it will determine which method makes sense for you.

Why doesn't QuickTime record internal audio on Mac?

QuickTime Player, the app that ships with every Mac, has a New Audio Recording command in its File menu. It captures your microphone. Only your microphone. That is not a missing feature — it is the design.

Apple's Core Audio architecture exposes microphone hardware as an input source to apps, but it does not expose "the audio currently playing on this Mac" as an input source. So when you launch QuickTime, choose File → New Audio Recording, pick your USB mic or built-in mic, and click Record, you get a clean track of your own voice. If a Zoom call is playing through your speakers at the same time, your recording will not include it. Your mic will pick up whatever bleeds into the room, but that is acoustic bleed, not system audio.

QuickTime's New Screen Recording is a different story. That command does capture system audio, because screen recording uses Apple's video pipeline which has access to audio streams. The catch: the output is a video file, not an audio file. You can extract the audio later, but it is two extra steps and an extra codec to wrangle.

What macOS actually routes (and what it hides)

Open Audio MIDI Setup — it lives in Applications/Utilities — and you will see a list of your hardware: built-in speakers, your USB mic, maybe a Thunderbolt audio interface. These are your physical endpoints. They are the only things apps can directly send audio to or receive audio from.

What you will not see, by default, is anything representing "the audio currently playing." That stream exists, technically — every sound goes through Apple's Core Audio HAL (hardware abstraction layer) before reaching your speakers. But the HAL exposes it as an output, not as something another app can subscribe to as an input. There is no menu toggle, no System Settings switch, no "expose system audio as an input" checkbox.

The reason for this is mostly about preventing apps from snooping on each other's audio without permission. macOS does eventually let authorized apps capture audio — but only through specific, narrowly-scoped frameworks. Core Audio, ScreenCaptureKit, and a handful of AVFoundation APIs are the gates. Outside of those, you need to either route audio yourself (with a virtual cable) or use an app that already has the right entitlements.

What is a loopback device, really?

A loopback device is a fake audio interface that macOS thinks is real. It has an input and an output. The trick: whatever audio goes into its output gets routed back to its input. So you set your Mac to play through the loopback device's output (instead of your speakers), and your recording app captures from the loopback device's input. The audio makes a round trip.

The most common way to do this is with a virtual audio driver. BlackHole is the open-source option — free, well-maintained, and quietly excellent. Install it once, restart your audio routing preferences, and BlackHole appears as a 2-channel or 16-channel audio device. You point apps at it for output, and they appear as inputs.

macOS also has a built-in concept called an Aggregate Device. You create one in Audio MIDI Setup by checking multiple physical devices into a single virtual device. This is not exactly loopback — it does not by itself route audio back to itself — but you can pair an aggregate with a Multi-Output Device so the same audio plays to your speakers and to a recorder. That is the classic DIY setup people used before BlackHole became standard.

Why does mixed loopback break down on calls?

Here is the actual problem loopback cannot solve: when you capture system audio during a Zoom, Meet, or Teams call, you are capturing the mix. Your voice (routed through the meeting app's echo cancellation) plus every other participant's voice, all collapsed into one stereo stream. That stream transcribes — Whisper handles mixed speech well enough — but you lose track of who said what.

For a one-off interview, a single mixed track might be fine. For a meeting you need to act on, or a customer call you need to summarize, mixed audio is a nightmare. Two people say "yes" at the same time. Someone cracks a joke while another is mid-sentence. The transcript has rows of unattributed quotes.

This is also why voice-to-text apps that only transcribe the system track leave you doing the "wait, who was that?" work yourself. The fix is per-track capture: separate your microphone input from the system audio before the recording starts, so your mic gets transcribed as one stream and the meeting audio gets transcribed as another. You still have to label speakers if you want names attached, but at least the streams are clean.

How ScreenCaptureKit changed Mac audio capture

Before macOS 13, capturing app audio required either a virtual audio driver or a recording app with special entitlements. Apple introduced ScreenCaptureKit — originally for screen recording — and steadily expanded it to cover audio. The framework exposes audio streams from individual apps, with the user's explicit consent.

For you, this means a class of newer apps can capture audio from a single app (say, Spotify) without affecting your microphone or system audio. Several paid utilities take advantage of these newer frameworks. And so does MetaWhisp, for a specific reason I will get to.

The privacy model is important. ScreenCaptureKit requires the user to grant Screen Recording and Microphone permission through System Settings → Privacy & Security. macOS shows a clear prompt the first time the app tries to capture. There is no silent background capture — Apple's whole point with this stack is that audio capture must be visible to the user.

What is per-track capture for meetings?

Per-track capture is the approach MetaWhisp takes for meeting audio. The idea is simple: instead of mixing your voice and the meeting audio into one stream, capture them as two separate audio tracks from the moment recording starts.

In practice, this works like this. When MetaWhisp detects a meeting window — Zoom, Google Meet, Microsoft Teams, FaceTime, Webex, Slack huddles, Discord — it opens two parallel capture streams. One reads from your microphone input. The other reads the system audio associated with the meeting app, using the entitlements and frameworks macOS provides for that. Both streams get timestamps; both are saved as separate files alongside your local transcript.

The local Whisper model handles your microphone track — that is the same path as normal voice-to-text dictation, free and offline, no account, no upload. The system audio track is opt-in: you turn it on explicitly in settings, and recording is OFF by default. When it is on, you can choose to keep system audio on-device too (the local Whisper model can transcribe both tracks) or send it to a transcription service you control via your own API key.

This split solves the "who said what" problem at the source rather than after the fact. Your mic track is your voice, attributed to you by default. The system audio track is everyone else — mixed, yes, but cleanly separated from your contribution, so you can read the transcript without your own lines muddying the output.

How to capture call audio with MetaWhisp

If you only need to capture your own voice during a meeting — for taking notes, for AI-cleaned summaries, for handing off the transcript to a colleague — you do not need anything special. MetaWhisp's global hotkey (default Right Option ⌥) works the same in a meeting as anywhere else. Hold the hotkey, talk, release, and the dictated text pastes wherever your cursor is. Your voice is transcribed locally, the audio never leaves your Mac.

If you want both sides captured as separate tracks, the steps are:

1. Install MetaWhisp on macOS 14 or 15, Apple Silicon. Free, no account.
2. Open the app once and grant the system permissions it requests: Microphone, and Screen Recording (the latter is what unlocks per-app audio capture).
3. Open Settings → Call Capture and toggle on Record meeting audio. This is OFF by default — you have to opt in.
4. Start your Zoom, Meet, Teams, FaceTime, Webex, Slack huddle, or Discord call. MetaWhisp detects the active meeting window automatically.
5. Use the global hotkey as usual to dictate. Your mic track is captured and transcribed locally on every press.
6. When the call ends, the system audio track is saved alongside your transcript in the local library. You can transcribe it locally too, or skip it if you only wanted your own side.

Local capture is free and unlimited. If you want AI post-processing — structured summaries, corrected transcripts, rewrites — you can plug in your own OpenAI or Cerebras API key in Processing Modes. Only the transcript text (never the audio) goes to your API endpoint, not to MetaWhisp's servers. If you would rather not manage a key, MetaWhisp Pro covers it with built-in cloud AI and adds cloud transcription for long meetings; see the pricing page for the current breakdown.

Pro tip: If you want clean transcripts from calls, get the system audio out of the room before you start. Use headphones, not speakers. Speakers bleed the meeting back into your microphone, and MetaWhisp captures your mic track only — so any speaker audio that bleeds into your mic will be transcribed as your words. Headphones make per-track capture work the way it should.

Which method should you pick?

Method	Captures system audio	Per-track separation	Setup complexity	Free
QuickTime (audio-only mode)	No — mic only	—	Trivial	Yes
QuickTime (screen recording)	Yes, but as a video file	No	Easy	Yes
Aggregate Device + Multi-Output	Yes	No (mixed stream)	Moderate	Yes
BlackHole loopback	Yes	No (mixed stream)	Moderate	Yes
ScreenCaptureKit-based paid apps (Loopback, etc.)	Yes	Some offer per-app capture	Easy	No
MetaWhisp (local model)	Yes, when toggled on	Yes — mic and system as separate tracks	Easy	Yes (local)

If your goal is a one-off system audio recording with no transcription needs — say, capturing a stream of audio for editing — BlackHole plus Audacity or QuickTime is the most flexible free path. If your goal is transcription of your own voice for dictation, MetaWhisp alone does that on-device without any of this routing. If your goal is a meeting transcript with both sides captured cleanly, per-track capture wins. For deeper coverage of meeting transcription specifically, see how to record and transcribe any call on Mac and the Zoom transcription walkthrough.

Frequently asked questions

Written by Andrew Dyuzhov — solo founder of MetaWhisp, marketer-turned-builder with ADHD who assembled MetaWhisp using AI coding tools on top of open-source Whisper. I dictate in Russian and English daily and built the app because nothing on the market did what I needed. I run an M-series Air and tested every method above on it. If you spot something wrong or out of date, find me on X.

• How to Record Voice on Mac (the mic-only path, no routing required)
• Record and Transcribe Any Call on Mac (per-app capture for the messy ones)
• Zoom Transcription on Mac (the Zoom-specific walkthrough)
• Download MetaWhisp — free, on-device, no account