How to transcribe audio to text on Mac (free and offline)

What macOS can transcribe without extra software

Apple has added transcription to several apps, but each one transcribes its own recordings. None of them has an import this MP3 button, which is the gap this guide fills.

• Voice Memos: Apple's guide says that with macOS Sequoia and a Mac with Apple silicon, speech in your recordings can be recognised and transcribed. Open a recording, click the Transcription button, then select text, Control-click and choose Copy. A recording made in macOS 14 or earlier, or on a device that doesn't support transcription, is transcribed automatically when you open it, if it contains speech. Apple says the feature isn't available in all countries or regions.
• Notes: audio recorded inside a note gets a transcript on Macs with M1 or later, when the device language is set to a supported one. Apple lists English (Australia, Canada, New Zealand, South Africa, UK, US), Spanish, Portuguese, Italian, French, German, Japanese, Korean, Simplified Chinese and Traditional Chinese. Double-click the recording, then use the More button to choose Copy Transcript or Add Transcript to Note. The guide says all features need iCloud notes.
• Live Captions: an accessibility feature that captions audio from all apps, or from the microphone, in a floating window. Apple says it needs a Mac with Apple silicon and that audio is processed on device. It's a live view; Apple's page doesn't describe saving the captions as a file, and it warns that accuracy may vary.

So if the audio is a Voice Memos recording, try Voice Memos first. If it's an MP3 or WAV from somewhere else, you need one of the two routes below.

Uploading the file to a hosted service

Pages such as Otter's MP3-to-text tool are the usual top results. They work, and the trade-offs are stated on their own pages. Otter's page says to log in or create a free account, then upload the file with the Import button or drag it into the Conversations tab. Its free plan is listed at 300 transcription minutes a month, files up to 90 minutes long, and 3 pre-recorded file imports per user, lifetime, according to the pricing page. Pro is $16.99 per user per month ($8.33 billed yearly) with 10 imports a month and 1,200 imported minutes. Transcription is offered in English (US and UK), Spanish and French, one language at a time.

What you give up is simple to state: the file leaves your Mac, the service keeps a copy of the transcript on its servers, and the limits are counted against your account. For a lecture you recorded yourself that's probably fine. For a client call, a medical consultation or an interview with someone who was promised discretion, it may not be.

Transcribing the file on the Mac itself

The third route runs a speech-recognition model on your own computer. Whisper, the open model behind most of these tools, comes in several sizes, and the smaller ones run comfortably on a recent Mac.

QuietNote is a note taker for Mac that does this. You import a common audio or video file, it's transcribed on your Mac with the Whisper model that ships with the app, and the result is a Markdown file with timestamps in a folder you choose. Nothing is uploaded and no account is needed. A summary with action items is optional: a local model through Ollama, fully offline, or Claude with your own API key, which sends only the transcript text and only when you ask.

1. Import the audio file into QuietNote.
2. Wait while it's transcribed on the Mac. Longer files take longer; how long depends on the Mac and the model size.
3. Open the Markdown transcript, or ask for a summary and action items if you've set up a local model or your own key.

The first signed Mac release isn't out yet; you can leave your email for it. QuietNote needs macOS 14.2 or later.

Which route fits

Getting a transcript you can use

Whatever tool you pick, the recording matters more than the software. A few practical points:

• Mono or stereo: speech recognition works on a single channel, so a stereo file is mixed down before it's recognised. That's fine for ordinary recordings. If you recorded two people onto separate left and right channels, splitting the file into two mono files first gives you a clean transcript per speaker.
• Background noise: fans, cafés and keyboard clatter cost accuracy. A microphone close to the speaker helps more than any setting afterwards.
• Speaker separation: a single mixed track has no reliable way to tell who said what. Whisper transcribes words, not identities, so expect to label speakers yourself on one-track recordings. Timestamps make that quicker.
• Accuracy: Whisper-class models do well with clear, close-miked speech in one language, and worse with crosstalk, heavy accents, jargon and names. Whisper's README notes its English-only model variants tend to perform better for English. Read the transcript before you quote it; don't trust a percentage you see on a landing page without a source.
• Time: on a recent Mac with Apple silicon, local transcription usually finishes in less time than the recording lasts, but it depends on the Mac and the model size. Otter's page says a five-minute file takes about five minutes.

What to do with the transcript

• Summary: a short paragraph of what was discussed. With a local model this stays on the Mac.
• Action items: who does what by when, lifted from the text rather than remembered afterwards.
• Quotes: pick the lines you'll reuse, with their timestamp so you can check the audio later.
• Search: a Markdown file is searchable with Spotlight or any editor, which is the point of having text instead of audio.

If the recording was a meeting, the meeting summary template gives the transcript a shape.