How to Transcribe Audio on a Mac

What macOS does and doesn't do

Dictation, in System Settings under Keyboard, turns live speech from a microphone into text in any text field. It's for speaking as you type, not for files: you can't point it at an MP3. Dictation vs transcription explains why they're separate jobs.

Apple has added transcription to some of its own apps in recent versions, for recordings made in those apps and in a limited set of languages. If you record in one of them, check whether a transcript option appears. For a file from somewhere else, such as a Zoom recording, a podcast episode, a camera clip or an interview from a recorder, you need one of the routes below.

Route 1: run a speech model on the Mac

OpenAI's Whisper is open source and runs locally. Two common ways on a Mac:

• whisper.cpp, a C/C++ port that runs efficiently on Apple silicon without Python.
• The original Python package, installed with pip, which needs ffmpeg.
• Third-party Mac apps that wrap these models in a graphical interface.

Local transcription keeps files on your computer and costs nothing per file. It takes setup, and larger, more accurate models are slow on older Macs. How to run Whisper locally walks through installation, model sizes and commands.

Route 2: upload to mydubly

A hosted service needs no setup and runs on fast hardware. With mydubly:

1. Open audio to text for audio files or video to text for videos.
2. Add the file. Videos stay on your Mac; the browser extracts the audio and sends only that.
3. Choose the spoken language, or another of 21 languages for a translation.
4. Save whichever outputs you need: the plain text, the version with timestamps, or the SRT and VTT subtitle files.

It costs 1 credit per minute, with 100 free credits on sign-in, for files up to 2 hours.

Route 3: dictation workaround (not recommended)

Some people play a recording through the speakers with dictation listening. It loses accuracy to room echo, stops on pauses, and ties up the Mac in real time. Either of the routes above is faster and more accurate.

Getting audio out of Mac apps

Voice Memos

Drag a memo to the desktop or use Share to save it; memos are M4A files, which upload directly.

QuickTime Player

Open a video and use File > Export As > Audio Only to save an M4A.

GarageBand

Share > Export Song to Disk writes MP3, AAC or uncompressed audio.

Photos

Export the video file, then upload it; the audio is extracted in the browser.

iMovie

File > Share > File exports the project; there's also an audio-only option in the format menu.

How to extract audio from video covers more tools if you want a small audio file before uploading.

Choosing a route

Privacy first, no budget

Local Whisper. Nothing leaves the Mac.

Speed and no setup

A hosted service. Upload, wait a few minutes, download.

A language you don't speak

A service that translates, or local Whisper's translate task, which outputs English only.

Subtitles for a video

Either; both produce SRT. A service gives VTT too.

Limits and trade-offs

• Local models on older Intel Macs can run slower than real time with large models; smaller models are faster but less accurate.
• Hosted services send audio off your Mac. With mydubly, audio and results are deleted within 30 minutes of delivery; cloud speech processing privacy lists questions for any service.
• Neither route labels speakers out of the box in mydubly's case; transcripts are plain text.
• Files longer than 2 hours need splitting for mydubly.

Recordings that start on an iPhone

Many Mac transcriptions start as iPhone recordings: Voice Memos, a screen recording, a video from the Camera app. Voice Memos sync to the Mac through iCloud when it's turned on, so the memo appears in the Mac's Voice Memos app ready to drag out. Videos come across with AirDrop or Photos. If you'd rather not move files at all, the same upload works from Safari on the iPhone itself; recording audio on a phone covers phone recording habits that make transcripts more accurate, such as holding the phone close to the speaker and avoiding wind.

A quick accuracy test

Before committing to a route for a big project, run the same 5-minute excerpt through each option you're considering and compare the results with the audio. Look at names, numbers and the noisiest passage. The tool that gets those right on your recordings is the one to use, whatever its reputation. Word error rate explained shows how to score the comparison if you want a number, and the how to transcribe audio guide covers preparing files so every tool does its best.