How to transcribe video to text on a Mac, offline

To transcribe video to text on a Mac without uploading it, use a tool that runs a speech model on the machine itself: a dedicated transcription app such as MacWhisper, the built-in transcription in Final Cut Pro, Premiere Pro or DaVinci Resolve Studio, or an editor such as cinemaOS, which transcribes with J.A.R.E.D. and keeps the transcript inside the project. Whichever you choose, the accuracy you end up with depends as much on three things you control as on the app: clean audio, the right language setting, and a short correction pass for names and speakers before the text goes anywhere else.

This guide covers the options fairly, then walks through a local workflow from video file to a corrected transcript and finished subtitles.

Why transcribe locally at all

Cloud transcription services are fast and often convenient, especially when several people need to comment on the same transcript. The trade-off is that your footage, or at least its audio, leaves your Mac and sits on someone else's servers under their terms.

For a lot of work that is fine. For some it is not:

• Interviews with people who agreed to be filmed, not to be processed by a third party. Documentary subjects and sources often fall here.
• Client material under NDA. Corporate, legal and medical recordings frequently come with rules about where files may go.
• Unreleased work. Embargoed films and products are easier to protect when the audio never travels.
• Poor or no connection. On location, on a train or on a set without Wi-Fi, local transcription simply works.

If privacy is the deciding factor, on-device AI video editing and privacy explains what local processing does and does not guarantee.

Your options on a Mac

There is no single best tool. It depends on whether you only need words, or words that feed an edit.

• A dedicated transcription app. MacWhisper runs open speech models locally, including OpenAI's Whisper, and exports common text and subtitle formats. It is the most direct route if you want a document and nothing else. Check the developer's site for current pricing and features.
• Open-source command-line tools. whisper.cpp and similar projects run the same family of models from Terminal. Free, flexible, and best suited to people comfortable with the command line.
• Final Cut Pro. Recent versions can transcribe speech into captions on Apple silicon Macs. Useful if you already cut in Final Cut and mainly need captions. Check Apple's documentation for the macOS and hardware requirements.
• Premiere Pro. Speech to Text and text-based editing use downloadable language packs, and transcription runs on the computer once a pack is installed. A natural choice if your work already lives in Creative Cloud; check Adobe's documentation for current language support.
• DaVinci Resolve Studio. Transcription and text-based editing have been part of the paid Studio version rather than the free one. A good fit if you finish in Resolve anyway; check Blackmagic Design's feature comparison for the current split.
• Cloud services. Descript, Otter and similar services process audio on their servers. Often very convenient, but read their data terms before uploading client material. Descript alternatives with local transcription compares the trade-offs in more detail.
• cinemaOS. An editor rather than a transcription app. Transcription runs on the Mac with bundled models: Parakeet for its published set of 25 languages, and a bundled Whisper route for other languages. The transcript stays in the project and drives text-based editing, search and subtitles. It is Mac-only, on Apple silicon.

Feature sets change between versions, so confirm language support and export options on each vendor's site before committing a project.

Before you press transcribe: fix the audio

Every speech model makes more mistakes on poor audio. A few minutes here saves far more time in correction.

• Use the best microphone track. If you recorded a lavalier and a camera mic, transcribe the lavalier. Camera audio picks up the room.
• Mute music and effects. Background music under dialogue is a common cause of missed or invented words. Transcribe the dialogue before the mix, not after.
• Split long recordings sensibly. A three-hour shoot is easier to check in sections, and a failed job costs less.
• Know the language. Auto detection usually works for single-language recordings. If speakers switch between languages, set the language yourself and expect to correct the switched passages by hand.

How to transcribe video to text with a local app

If you only need the words, the steps are much the same in most local transcription apps:

1. Open the video file. Many apps accept video directly and read its audio track. Command-line tools often expect an audio file instead, so you may need to extract the audio first, for example with ffmpeg.
2. Set the language. Choose it yourself when you know it. Auto detection works on single-language material but can guess wrong on short or mixed clips.
3. Pick a model size. With Whisper-based tools, larger models are generally more accurate and slower. Try a short clip with two sizes before committing a long recording.
4. Run the job and correct the text. Use the checklist further down for names, numbers and quiet passages.
5. Export what you need. Plain text for reading and quoting, SRT or WebVTT when you want timecodes or subtitles. Export options vary by app, so check yours.

If the transcript is going to feed a cut or subtitles, working inside an editor saves a round trip of exporting and re-importing, which is what the next section shows.

How to transcribe video to text in cinemaOS

cinemaOS is a Mac-native editor for Apple silicon. Its transcription is part of the editing workflow rather than a separate app, which matters if the transcript is going to become a cut or a set of subtitles. The transcribe and edit by text tutorial shows the transcribe and edit steps in about a minute.

Step 1: Transcribe with J.A.R.E.D.

Import the video into a project and select the clips or scene. Press ⇧⌘T, Transcribe with J.A.R.E.D. The job runs locally, with no generative credits and no upload. You can also start from the Finisher in Localize ▸ Captions, choosing Auto Detect for the spoken language.

On an 8 GB Mac, heavy AI jobs run one at a time, so start a long transcription before a break rather than alongside other model work.

Step 2: Correct names and speakers

cinemaOS assigns up to four speakers per transcription. How well it separates them depends on the recording: distinct voices on separate mics usually separate well, while overlapping speech and short interjections are where assignments most often go wrong.

Work in this order:

1. Name the speakers. Replace generic labels with real names. Search and reading both get faster.
2. Fix proper nouns. Names, brands, places and technical terms are the usual misses. Correct each one once and search the transcript for other instances of the same mistake.
3. Check speaker changes that matter. Fix the turns that affect meaning or captions. A misattributed "right" in the middle of an answer can usually wait.
4. Leave the rest for later. If the transcript is feeding an edit, proofread only the lines that survive the cut.

For more than four speakers, such as a panel, name the four main voices and correct the others by hand.

Step 3: Use the transcript in the edit

Open the Text Editing tab in the Sequencer bin. Selecting words selects the matching media, and removing a passage cuts the linked picture and sound. Which edits are available depends on the source material and its transcript, so check the timeline after each one. Use Search Project… to find exact words across every transcribed clip, which is the quickest way to locate a line you remember hearing.

Treat the transcript as a map. Play every join and listen for clipped words before you move on. Transcribe an interview and edit by text covers the editorial side of this in depth.

Step 4: Export text or subtitles

The corrected transcript becomes your caption source. In Localize ▸ Captions, review timing and line length, then export the format your destination expects: SRT, WebVTT, ASS, SCC, EBU-STL, Apple iTT or FCPXML captions. You can also burn subtitles into the picture for social versions, or translate them on the Mac, as the offline subtitles tutorial shows.

If you need a plain document rather than subtitles, export SRT or WebVTT. Both are plain text files that open in TextEdit or any word processor, with timecodes you can keep for reference or remove. If a formatted document is all you ever need, a dedicated transcription app is the more direct route.

A correction checklist that works with any tool

Whatever produced the transcript, the same pass catches most problems:

• Search for every name in your call sheet or contributor list and check each spelling.
• Read numbers aloud. Dates, prices and quantities are often transcribed in an inconsistent form.
• Check the first and last lines of each clip. Models sometimes miss words that start or end abruptly.
• Listen to the quiet passages. Low-level speech is where words go missing, and some models fill silence with text nobody said.
• Watch for confident nonsense. A sentence that reads smoothly but does not match what you remember hearing deserves a listen.
• Keep one corrected master. Fix the transcript before it becomes subtitles or a translation, so the same error does not spread into every version.

What cinemaOS does not do

Being clear about limits saves you a wasted afternoon:

• It is Mac-only, on Apple silicon. Windows is planned for November 2027.
• Speaker separation is limited to four voices per transcription, and quality depends on the recording.
• It does not export a formatted transcript document such as a Word file. Its transcript exports are timed-text and subtitle formats.
• It is not a collaborative cloud transcript tool. There is no shared online document for a team to edit at the same time.
• It does not do lip sync. If you go on to dub the film, mouth movements are not altered to match the new language.

FAQ

Can I transcribe video to text on a Mac for free?

Yes. Open-source tools such as whisper.cpp run locally at no cost, and some transcription apps have free tiers. cinemaOS includes transcription in its plan, which costs $9.99 USD per month excluding VAT after a 60-day free trial with no card and no automatic charge.

How accurate is offline transcription?

Current local models handle clean, single-speaker audio well. Accuracy drops with music under dialogue, heavy accents, crosstalk and specialist vocabulary on any tool, so plan a correction pass for names and terms.

Does cinemaOS upload my video to transcribe it?

No. Transcription with J.A.R.E.D. runs on the Mac with bundled models. The video, audio and transcript stay in your local project.

Start with one clip

Pick a short interview with a few names in it, transcribe it locally in one or two of the tools above, and time how long the correction pass takes. That tells you more than any feature list. If you want the transcript, the edit and the subtitles in one project on an Apple silicon Mac, cinemaOS is free for 60 days with no card and no automatic charge.