Guide

Offline transcription: every way to transcribe audio without uploading it

Most transcription services want your recording on their server. Here is every way to keep it on your own Mac or iPhone, including the free ones and the ones that are not ours.

9 October 2026

Search for "transcribe audio" and nearly every result is a website that asks you to upload the file. For a podcast episode that is fine. For a client call, a source interview, a therapy session or a board meeting, it means a copy of the recording now sits on someone else's server, under a retention policy you cannot inspect.

You do not have to accept that. Speech recognition models small enough to run on a laptop have been good for several years now, and there are more ways to use them than most people realise. This guide covers all of them, including the free ones and the ones that compete with us.

Why "offline" is worth caring about #

Three practical reasons, beyond principle.

Confidentiality. If a recording is covered by a consent form, an NDA, legal privilege or health privacy rules, uploading it to a third party is a decision with consequences. Processing it locally removes the question entirely.

Cost. Cloud services pay for server time, so they meter you — by the minute, the hour, or a monthly allowance. Local transcription runs on hardware you already own, so a two-hour file costs the same as a two-minute one.

It works anywhere. On a plane, in a courthouse with no Wi-Fi, on a field recording trip. No connection, no problem.

One caveat applies to everything below: offline is a claim you can test rather than trust. Our guide on checking whether a transcription app uploads your audio shows how, in about thirty seconds.

Option 1: What is already on your iPhone or Mac #

Voice Memos and Notes transcripts. Recent versions of iOS can show a transcript for a voice memo, and call recordings in Notes come with one. This runs on the device and costs nothing.

It is the right answer if you record on your iPhone, you speak one of the supported languages, and you only need to read the text. The limits are real, though: the language list is short compared with Whisper's, it does not work on files recorded elsewhere, there is no subtitle export, and the transcript is one long block of text without timestamps you can use. Our guide on transcribing a voice memo on iPhone covers it step by step.

Dictation. Dictation on a Mac or iPhone turns live speech into text, and on recent hardware much of it runs on the device. But it listens to the microphone in real time. It cannot read an audio file, so it is the wrong tool for anything you have already recorded.

Option 2: Run Whisper yourself #

OpenAI released Whisper as open source, and two projects make it easy to run on your own machine.

whisper.cpp is a fast C/C++ port that runs well on Apple silicon. It is free, it is excellent, and it is what Offline Transcription is built on. You use it from Terminal: compile it, download a model, convert your audio to the format it expects, and run a command. The output is accurate and you control every setting.

The original Python version from OpenAI does the same job with a heavier setup — Python, a package manager, and ffmpeg.

Choose this if you are comfortable in Terminal, you want the largest models, or you want to script transcription into another workflow. The cost is your time: installing, converting formats, and cleaning up the output yourself. We wrote a frank comparison of running whisper.cpp yourself versus using an app.

Option 3: Free desktop apps #

Buzz is an open-source desktop app that puts a window around Whisper. It runs on macOS, Windows and Linux, takes audio and video files, and exports subtitles. The open-source build is free to download; a paid version is also available on the Mac App Store.

It is the obvious choice if you want a free graphical app, or you need the same tool on Windows. Expect an interface that is functional rather than polished, and check the project page for the current release, since behaviour on very long files varies between versions.

Option 4: Paid Mac and iPhone apps #

Several native apps package Whisper for Apple devices. All of the ones below process audio locally.

MacWhisper is the most feature-rich. Its Pro tier runs the largest Whisper models, labels speakers, records system audio and takes YouTube links. It also costs several times more than the simpler apps. Our MacWhisper comparison goes through when it is worth it.

Aiko, by Sindre Sorhus, runs on Mac and iPhone, uses large models on the Mac, and keeps the interface minimal. It favours accuracy over speed.

Offline Transcription — ours — is the simple, inexpensive option. Drop in one file or a folder, transcribe, and export to TXT, SRT, SBV, VTT, CSV, HTML or PDF. It runs on Mac, iPhone and iPad with one purchase, transcription itself is free and unlimited, and Pro is $12.99 once. It ships the Tiny, Base and Small models, and it does not label speakers or record audio.

Which one should you use? #

If you…	Use
Recorded on your iPhone and just want to read it	The built-in Voice Memos transcript
Want free, and are comfortable in Terminal	whisper.cpp
Want free, with a window, or need Windows	Buzz
Have difficult audio and need speaker labels	MacWhisper Pro
Want something simple for regular files, on Mac and iPhone	Offline Transcription

The honest tiebreaker is your own audio. Every option here is free to try, and none of them uploads anything, so run your hardest recording — the noisy one, the one with the strong accent — through two of them and compare. Our guide on improving transcription accuracy explains why the recording usually matters more than the app.

What you give up by staying offline #

Being fair to the cloud services: they do have advantages.

Bigger models on bigger hardware. A server can run the largest models quickly. On a laptop the large models are slow, which is why most local apps default to smaller ones. On clean audio the difference is small; on difficult audio it is noticeable. Our guide to Whisper model sizes covers where the line falls.

Collaboration features. Shared workspaces, comment threads and automatic meeting bots are built around the recording living on a server. A local app will not join your Zoom call for you.

Speed on older machines. On an Intel Mac, local transcription is meaningfully slower than on Apple silicon. Our guide on why transcription is slow explains why.

If none of those matter for a given recording, there is no reason to upload it. And if the recording is confidential, they should not matter.

The short version #

You can transcribe almost anything without it leaving your device. If you record on an iPhone, start with the built-in transcript. If you want free and flexible, use whisper.cpp or Buzz. If you want an app that just works on your files, pick one of the native apps — and test it with Wi-Fi off before you trust it with anything sensitive.