How to transcribe video offline on a Mac

The short answer

Use an app or tool that runs a speech-recognition model on your own machine, most commonly one based on the open-source Whisper models. You download the model once, then transcription happens entirely on your Mac. No account, no per-minute billing, and the recording never leaves your computer.

Why offline transcription matters

Cloud transcription services are convenient, but they require uploading the file. For interviews with sources, client calls, medical or legal material, or simply family videos, that is often a deal-breaker. It can also be slow on large files and gets expensive when priced by the minute.

Local transcription flips the trade-off: the cost is some processing time and disk space for the model, and in return nothing leaves the machine. On Apple Silicon, speech models can use the GPU, which makes long files practical.

Your options on a Mac

macOS dictation is designed for speaking into a text field, not for transcribing a file, so it isn't the right tool here. The realistic choices are:

• Command-line tools. Open-source projects such as whisper.cpp are free and run fully offline. You'll need to be comfortable with the Terminal, download a model file yourself, and usually convert video to audio with ffmpeg first.
• Transcription apps. Several Mac apps wrap Whisper in a friendly interface. They're good for one file at a time; check whether they keep your transcripts searchable afterward or just hand you a text file.
• A searchable library. If you'll transcribe many recordings, an app that also indexes the results lets you search across all of them later instead of opening transcripts one by one.

Picking a model size

Whisper-style models come in sizes. Smaller models are quick and fine for clear speech; larger ones handle accents, noise and technical vocabulary better but need more memory and time. A sensible approach is to start mid-sized, check a few minutes of output, and only step up if you see errors that matter to you. Whatever the size, clean audio helps more than anything else: a close microphone beats a bigger model.

How to do it in MediaFind

MediaFind transcribes with Whisper on your Mac, keeps word-level timestamps, and makes every transcript searchable.

1. Download MediaFind for macOS (Apple Silicon, macOS 14 or later). It's signed and notarized.
2. On first run, choose a transcription model that fits your machine, speed or accuracy. A Turbo tier brings near-Large accuracy to 8 GB Macs, and on Apple Silicon you can switch to the whisper.cpp engine, which runs on the GPU.
3. Add the folder with your videos or audio. The model downloads once; after that you can disconnect from the internet.
4. Add names and terms your recordings use (people, products, jargon) so they're spelled correctly.
5. Open any file to read its transcript with speaker labels, or search across everything at once.
6. Export the transcript as Word, PDF, SRT, VTT or plain text. The export is written on your machine.

Tips and limitations

The first run of a big folder is the slow part. Let it work through the backlog; files become searchable as each one finishes, and new files you add to a watched folder are picked up while the app is open.

Local transcription isn't magic: overlapping speakers, loud music and poor microphones still cause mistakes, and you should proofread anything you plan to quote or publish. If you need subtitles specifically, see making SRT subtitles locally. For background on how the engines compare, read on-device transcription.

Frequently asked questions

Can I transcribe video on a Mac with no internet connection?

Yes, once the speech model is downloaded. Local tools and apps like MediaFind run the model on your Mac, so transcription works fully offline after that one-time download.

Does built-in macOS dictation transcribe video files?

Not really. Dictation is meant for speaking into text fields. For recorded files you need a transcription tool that reads the audio track directly.

Do I need an Apple Silicon Mac?

For MediaFind, yes: the Mac app requires Apple Silicon and macOS 14 or later. Some command-line tools also run on older Intel Macs, though more slowly.

Does MediaFind work on Windows or Linux?

Windows and Linux builds are available in beta. The Mac app is the main, signed and notarized build.