How to transcribe an audio file you already have on a Mac

Start with the disappointing part: Narada cannot do this. It records meetings — your microphone and your Mac’s system output, captured live — and transcribes those two files on the machine afterwards. There is no “open file” menu item, no watch folder, no import. If you already have the audio, Narada is the wrong shape of tool and no amount of wanting it to work will change that.

So here is the guide for the tools that do fit, ranked by how little of your time they cost, with prices and limits quoted from the vendors' own pages.

What Apple gives you, and exactly where it stops

Two Apple apps transcribe audio, and both are recorders that happen to transcribe rather than transcribers that happen to record.

Voice Memos. Apple’s requirement sentence: “With macOS 15 or later and a Mac with Apple silicon, speech in your audio recordings can be recognized and transcribed to text in Voice Memos.” Old files are handled too — “If you open a recording made in macOS 14 or earlier, or a recording made on a device that doesn’t support transcription, Voice Memos transcribes the recording automatically if it includes recorded speech” — which is the closest Apple comes to describing a back catalogue. There is a hedge attached: “Audio transcription isn’t available in all countries or regions.”

Notes. “Audio transcription is available on Mac computers with M1 or later when the device language is set to English, Spanish, Portuguese, Italian, French, German, Japanese, Korean, Simplified Chinese, and Traditional Chinese.” Ten languages, tied to your device language rather than to the language in the recording. No Russian, no Ukrainian, no Polish, no Dutch. A summary on top of the transcript needs Apple Intelligence switched on, which carries its own narrower locale list.

Now the part that decides this page. Neither Apple page documents bringing in a file from anywhere else. The Voice Memos article says nothing about importing; the Notes article covers recording inside a note and nothing beyond it. That is an absence of documentation, not a documented impossibility — but it means the Apple route for a file that arrived by AirDrop, Slack or SD card is undefined, and the workaround people fall back on (play it out loud, record it in Voice Memos) costs you the length of the recording in real time and degrades the audio on the way through.

MacWhisper — the shortest route for a file on disk

MacWhisper’s own pitch for this exact job: “Transcribe any file. Just drag and drop.” It names the formats — “mp3, wav, m4a, ogg, opus, mov and mp4” — so a video file works as well as audio, and claims “100 different languages”, which is a genuinely wider net than Apple’s ten or Narada’s twenty-five.

Pricing from the vendor’s table: “MacWhisper, Free, €0, / month, Free forever” and “MacWhisper Pro, Pay Once, €64, / license, Includes lifetime updates”. Free transcribes files. Pro is what you pay for when you want the surrounding machinery: batch transcription across many files, automatic speaker recognition, and exports beyond plain .txt — .md, .pdf, .html, .docx.

Privacy, in their words: “No data (audio, text or other) leaves your device.” Read that as scoped to local transcription, because their docs list the exceptions plainly — cloud transcription providers, DeepL translation and AI prompts each send data out, and on their own cloud transcription they write: “Note that any media files you transcribe with the Cloud transcription feature will be sent to our servers for transcribing, so don’t use this feature with sensitive data.” That sentence is more honest than most privacy pages manage.

One trap worth knowing before you buy: there are two products with almost the same name. The direct download from macwhisper.com is the one-time licence described above. The Mac App Store build is a different app called “Whisper Transcription”, it is a subscription, and per MacWhisper’s own docs it does not include automatic meeting detection. If you compare prices across a review site, check which one it was looking at.

Fuller comparison with Narada here.

Superwhisper — if you also want dictation everywhere

Superwhisper is a dictation tool first — hold a hotkey, speak into any app — with file handling attached. Their docs say it “can transform existing audio and video files into text”.

Prices are $8.49/month, $84.99/year, or $249.99 once, and their docs say “All plans include the same features and work on all platforms. The only difference is billing frequency.” New users get “15 minutes of free access to all pro features”.

The important asterisk for a privacy-driven choice: "Local voice models" are listed as a Pro feature. The free tier runs on cloud models, so “it works offline” is a property of the paid configuration, not of the app in general. Their privacy policy phrases it as design intent rather than a guarantee — “Superwhisper is designed to process audio data locally on your device” — and separately notes that speaker separation is off by default and also Pro.

Pick this one if you want dictation across your whole system and file transcription is a bonus. If you only ever have files, you are paying a subscription for a hotkey you will not use.

Aiko — one price, no surface area

The one product in this set that is genuinely file-in, text-out. $24 on the App Store, no in-app purchases, and the App Privacy section reads “The developer does not collect any data from this app.”

Its description does something rare and states its own gaps: “The transcription is powered by OpenAI’s Whisper running locally on your device. Nothing leaves your device.” Then: “It does not do live transcription while recording. It does not currently have speaker detection.” Audio in 100 languages.

No meeting capture, no calendar, no summaries. The vendor’s own instruction for a call is to record it in the conferencing tool and drop the resulting file in afterwards. If you want a box you drag a file onto once a month, this is the cheapest honest answer and the least likely to grow features you did not ask for.

whisper.cpp — the terminal route

Free, scriptable, and the right answer for four hundred files rather than one. It is also the only option here where you will meet the audio format requirement head-on.

Build it:

cmake -B build
cmake --build build -j --config Release

Fetch a model:

sh ./models/download-ggml-model.sh base.en

Then the constraint that catches everyone. The CLI “currently runs only with 16-bit WAV files”, so your m4a or mp3 has to be converted first, and the project gives you the line:

ffmpeg -i input.mp3 -ar 16000 -ac 1 -c:a pcm_s16le output.wav

On Apple Silicon “the inference runs fully on the GPU via Metal”, and there is an optional Core ML path that runs the encoder on the Neural Engine for “x3 faster compared with CPU-only execution” — at the cost of a Python toolchain (ane_transformers, openai-whisper, coremltools) and a -DWHISPER_COREML=1 build.

What you are trading: an afternoon of setup, no interface, no speaker labels out of the box, and you own the pipeline forever. What you get: zero cost per hour of audio, complete control of the model choice, and something you can put in a shell loop.

Which one, by constraint

When a recorder is the right tool instead

Transcribing a file you already hold is a rescue operation. The recording exists, whatever it captured is what you get, and every tool above is doing forensics on one mixed waveform. That is why speaker labels are the weakest column in the table: a single track containing four people is one signal, and separating it is guesswork that vendors gate behind a Pro tier because it is hard and it is wrong sometimes.

A recorder gets to solve a different, easier problem, because it is there when the audio is created. Narada opens two independent streams — the microphone and the Mac’s system output — writes them as separate files, and transcribes each. Your lines are attributed to you because they are in your own file, not guessed at from a mix. The other side’s audio gets speaker separation into Speaker 1, Speaker 2 and so on.

The honest limits of that, since this page has already spent its credibility on other people’s caveats:

• The labels have no names. Narada will not guess which cluster is which person, and separation exists only for the call’s audio. An in-person or mic-only recording gets no speaker split at all — everything is attributed to you.
• Transcription covers 25 European languages, unevenly. The engine is Parakeet TDT v3 through Core ML, and its published benchmark runs from about 4% word error rate on Italian to about 37% on Greek, with Slovenian, Latvian, Maltese and Lithuanian in the mid-twenties. Chinese, Japanese, Korean, Arabic, Hindi and Turkish are not supported. MacWhisper’s Whisper models cover far more languages than this.
• No import, no live transcription, no translation. Transcription runs after the recording stops. macOS 15.6 or newer, Mac only.
• macOS audio can fail mid-recording, on any tool. A device reshaped under a live capture or an overloaded audio daemon can cost you minutes. Narada compares each finished recording against itself and marks it Incomplete with a reason — it cannot recover audio that was never written, and neither can anything else.
• On-device AI summaries are Apple Silicon only, a one-time 3 GB or 7 GB model download depending on RAM, lighter than the cloud route, and paused while a recording is running. The recorder tier is free; Pro AI is $6.99/month or $69.99/year with a 3-day trial.

And the part no software covers: if the audio has another person in it, recording it may need their agreement, and that varies by country and by state. Nothing on this page makes a recording lawful — not the local ones, not the cloud ones. If the conversation is still ahead of you, say out loud that you are recording.

Sources

All fetched 4 August 2026.

• Apple Support, View a transcription of a recording in Voice Memos on Mac — macOS 15 + Apple silicon requirement, older-recording behaviour, region caveat.
• Apple Support, Record and transcribe audio in Notes on Mac — M1-or-later requirement and the ten device languages.
• Apple Support, Record audio in QuickTime Player on Mac — the single microphone input control.
• macwhisper.com — “Transcribe any file. Just drag and drop”, format list, +100 languages, €0 / €64 pricing table.
• docs.macwhisper.com — keeping transcriptions private and Assistant — local by default, the cloud exceptions, the sensitive-data warning.
• macwhisper.com privacy policy — “No data (audio, text or other) leaves your device.”
• docs.macwhisper.com — MacWhisper & Whisper Transcription difference — the App Store build is a subscription and lacks meeting detection.
• superwhisper.com/docs — introduction — “can transform existing audio and video files into text”; Pro — prices, local voice models as a Pro feature; privacy; Meeting mode — speaker separation off by default.
• Aiko on the App Store — $24.00, no in-app purchases, local Whisper, no live transcription, no speaker detection, 100 languages.
• github.com/ggml-org/whisper.cpp — build commands, model download script, “currently runs only with 16-bit WAV files”, the ffmpeg conversion line, Metal and Core ML notes.
• Narada: prices, the 3-day trial and the macOS 15.6 minimum from its App Store listing. Word error rates per language from the FLEURS benchmark table published in the documentation of FluidAudio 0.13.4, the transcription package Narada ships. https://raw.githubusercontent.com/FluidInference/FluidAudio/main/Documentation/Benchmarks.md