Questions about recording and transcribing calls on a Mac

Local runs the speech model on your own machine, so the audio never leaves it. Cloud sends the audio to a service that transcribes it and keeps a copy. The finished transcript can read the same either way; what differs is where the audio goes and what the service can build around it.

Yes, when the local tool keeps the audio on the machine. There's no copy on a server to be breached, handed over, or used to train a model. A cloud service holds your audio and transcript on its side, under its policies.

Not reliably, and it depends on the specific models more than on where they run. A cloud service can improve its model server-side without you doing anything; a local tool improves when you update the app. Neither guarantees a clean transcript of names or heavy accents.

Usually. It runs on hardware you already own, so there's no per-minute fee and no subscription that grows with your call volume. Cloud note-takers tend to charge once you pass a free tier.

Not on its own. Shared workspaces, search across everyone's meetings, and a bot that joins calls you're not on are cloud features. If a team needs one searchable record of every call, cloud is the better fit.

After the one-time model download, yes. The transcription runs with the network off.

Paste the full transcript into a model like ChatGPT or Claude and ask for a short summary, the decisions made, action items with owners, and any open questions. A specific prompt and a transcript that labels the speakers get you a usable answer instead of a vague recap.

Ask for named parts rather than a summary in general. Something like "Give me a two-line summary, then the decisions, the action items with who owns each, and the open questions" works far better than "summarize this".

It helps a lot. A summary is only as useful as knowing who agreed to what, so a transcript that separates the speakers lets the AI attribute decisions and action items to the right person.

The summary step sends the transcript text to whichever service you pick, so treat it like any other upload. If the conversation is sensitive, keep the recording and transcription local and decide per transcript whether the text should leave your machine at all.

Yes. Once you have the transcript as text, it does not matter which app the call was in. The same paste-and-prompt approach works for a Teams, Zoom, Meet or FaceTime call.

For the summary itself, a general model like ChatGPT or Claude is enough. The part worth solving separately is getting a clean, private transcript in the first place, which is a recording and transcription job, not a summarising one.

No. It records the audio that reaches your Mac, so your guest just joins the call as usual on whatever app you both use.

Not really. The WAV files are 16 kHz mono, made for transcription. For release-quality audio, record with a dedicated podcast tool or your usual rig, and use this for the transcript.

Yes. Your microphone and the far end are recorded as two channels, and a second guest on the far end is split into separate speakers after the interview.

No. Transcription runs locally on a downloaded model. The only network use is the one-time model download and the update check against GitHub.

macOS 14.2 or newer, where Apple added the audio tap that makes recording another app's sound possible. Apple Silicon is recommended.

The transcript is written and flushed after every segment, so everything up to the last line is on disk. On the next launch the app offers to finish the recording.

Speech-to-text where the model runs on your own computer instead of a cloud server, so the audio is turned into text without being uploaded anywhere.

The audio and the transcript stay on your machine, and after the one-time model download the recognition runs offline. Nothing is sent to a transcription service.

A recent Mac helps. The models run best on Apple Silicon, and the first run downloads about a gigabyte of model files; after that the work happens on your machine.

Most transcribe files you already have. Recording both sides of a call and transcribing it live is a narrower case, and it needs an app that captures system audio and your microphone as separate channels.

Yes. For a shared team workspace, search across many calls, or a synced mobile app, cloud tools do more. Local wins on privacy and cost, cloud wins on scale and collaboration.

After the model is downloaded once, yes. The recognition needs no connection; only the initial model download and, in some apps, an update check use the network.

Transcription is what was said. Diarization is who said it, splitting the audio into distinct speakers before anyone attaches real names.

No. It groups voices into "Speaker 1", "Speaker 2" and so on. Names come from you or from a meeting platform, not from the sound itself.

It removes the biggest split, you versus the far end, for free. If several people share the far-end channel, diarization still has to separate them.

It's the hard case. When two people talk at once on the same channel, any diarizer struggles to assign every word to the right voice.

In Transcriber it runs over the remote channel after the call by default. An experimental setting can do it live during the call, less accurately.

No. Since macOS 14.2 an app can record system audio through Apple's Core Audio process tap after you grant permission once. There is no driver to install and nothing to reroute.

The built-in screen recorder and QuickTime capture your microphone, not the sound your Mac plays. Recording the output needs an app that uses the system audio tap, or on older macOS a virtual audio driver.

It can. An app that records both keeps them as separate channels, so your voice and the other side stay on different tracks and can be labelled.

Not when the app transcribes locally. Transcription runs on your machine with a downloaded model, and the only network access is that one-time download and the daily update check.

macOS 14.2 or newer, because that is where the Core Audio process tap arrived. An Apple Silicon Mac is recommended for the transcription models.

Nothing joins the call, so there is no indicator on their side. Recording a conversation without consent is illegal in many places, so tell them before you start.

Not for a local recording. Meet's own recording is limited to paid Workspace editions (Business Standard and up, as of September 2026), but recording the call on your own Mac doesn't touch that at all.

Yes. The other people's voices come out of your browser as system audio, and that's captured the same way whether Meet is in Chrome, Safari, or a standalone window.

Yes. The far end arrives through your Mac's audio output as its own channel, so their words land in the transcript with a separate speaker label from yours.

No. Transcription runs on your machine with a downloaded model. The only network access is the one-time model download and the update check against GitHub.

It depends on where you and the other people are. In many places, including Germany under section 201 StGB, recording without consent is a crime. Tell everyone before you start.

Only a one-to-one FaceTime audio call, on recent macOS and where your region allows it, and it announces itself to both people. A video call or a group call has no built-in record button, so you record the microphone and the system audio yourself.

Yes. Their voice comes out of your Mac's speakers as system audio, so it's captured as its own channel and lands in the transcript with a separate speaker label.

Not automatically when you record locally. FaceTime's own audio recording plays a spoken notice, but a local recorder can't do that for you, so telling people is your job.

No. The transcription runs on your machine with a downloaded model. The only network access is the one-time model download and the update check against GitHub.

macOS 14.2 or newer, because that's where Apple added the audio tap that lets one app record another app's sound. An Apple Silicon Mac is recommended.

It depends on where each person is. In many places, including Germany, recording without consent is a crime. Tell everyone before you start.

With every app that plays audio through the Mac. Zoom, Teams, Google Meet, WhatsApp, FaceTime, a call in a browser tab and a YouTube video are all the same to it.

Microphone for your side and System Audio Recording for the other side. Without the second one, the far end of the call would be missing from the transcript.

About one gigabyte, downloaded once on the first recording or up front in Settings. It contains Parakeet for transcription, Silero for voice detection and pyannote for speaker separation.

The transcript is written to disk after every segment, so everything up to the crash is already saved. On the next launch Transcriber offers to finish the session.

Yes. Markdown is always written, JSON and SRT are optional exports in Settings, and the raw audio can be kept as WAV files.

macOS 14.2 is the requirement. The transcription models run much better on Apple Silicon, so that is what I recommend.

Not from the software side. Transcriber records what your Mac plays and what your microphone hears, so nothing joins the meeting and no icon appears for anyone else. Tell them anyway, recording without consent is illegal in many places.

Yes. Nothing on the meeting side has to be enabled or allowed. If audio reaches your speakers and your microphone, it can be recorded.

Any app that plays audio on your Mac. Zoom, Teams, Google Meet, WhatsApp, FaceTime and calls in a browser all work the same way.

No. Transcription runs on your Mac with a downloaded model. The only network access is the one-time model download and the daily update check against GitHub.

No. It records the audio on your Mac, so you have to be in the call. That's the price of not having a bot.

macOS 14.2 or newer, because that's where the system audio tap arrived. An Apple Silicon Mac is recommended for the transcription models.

No. Recording on your own Mac doesn't use Teams' recording feature, so an admin policy that blocks the in-app button doesn't apply. Consent from the other participants is still your responsibility.

Yes. Teams in a browser tab plays its audio through the Mac like any app, so the system audio tap captures it the same way it captures the desktop app.

Yes. The far end comes in through your Mac's audio output as its own channel, so their words land in the transcript with separate speaker labels.

No. Transcription runs locally with a downloaded model. The only network access is the one-time model download and the app's update check against GitHub.

macOS 14.2 or newer, because that's where Apple added the audio tap that makes recording another app's sound possible. Apple Silicon is recommended.

No. Transcriber listens to what your Mac plays and what your microphone hears, so it doesn't matter whether Zoom's own recording is on, off, or locked by the host.

Yes. The far end comes in through your Mac's audio output and is captured as its own channel, so their words land in the transcript with a separate speaker label.

No. Transcription runs on your machine with a downloaded model. The only network access is the one-time download of that model and the update check against GitHub.

macOS 14.2 or newer, because that's where Apple added the audio tap that makes recording another app's sound possible. An Apple Silicon Mac is recommended.

That depends on where you and the other people are. In many places, including Germany, recording without consent is a crime. Tell everyone before you press record.

No. Recording and transcription both happen on your Mac. The audio files stay in the folder you chose, or are deleted after the call if you turned off keeping audio.

Two. The one-time download of the transcription models, and a check for new versions against GitHub, once a day by default. Both can be seen in the app log.

No account, no login, no email address. Download the app, grant the two permissions, record.

In a folder you pick in Settings, one subfolder per recording. Your sync tool decides where they go from there, the app does not upload anything.

The app itself sends no personal data anywhere, which removes the usual processor question. Recording a conversation still needs a legal basis and the consent of the participants, that part is yours.

Yes, the source is on GitHub, so every claim on this page can be checked against the code.