What you’ll make
Save a checked plain-text transcript from a video on your Mac.
Worked example
Caption an interview for an oral history archive
You have a 55-minute interview that you can transcribe. One speaker is quiet, and the test clip includes the surname “Quintero.” You check the sample against the audio before processing the full recording.
You save a text copy people can search and an SRT for playback. You mark an unclear word at 43:12 and adjust a cue near the final question.
Before you start
- Start with a copy of a video you may transcribe, a short sample with both clear and difficult speech, and a list of terms worth checking. Decide which format your destination expects, such as SRT for subtitles.
- Open About This Mac to check your chip and macOS version. Confirm your chosen app or Homebrew package supports that setup, and leave enough disk space for the downloaded model and working files.
Find a short clip you can listen back to
Use a video you’re allowed to transcribe. Pick a 20 to 30 second excerpt with a name, a quiet speaker, or some background noise. Choose a Mac transcription tool with local processing and plain-text export that works on your Mac.
I have a 55-minute oral history interview. I pick the part where a quiet speaker says “Quintero.”
That gives me one difficult word to check before I run the whole interview.
Run a short local test
In MacWhisper, select a local model and set the spoken language, then import a copy of the short clip. Play it back while reading the result. If you use the Terminal workflow, follow the command example in the worksheet for your clip.
I run the excerpt with the same language setting I plan to use on the whole video.
I expect the quiet speaker’s sentence to be there. If I can hear it but the transcript drops it, I’d try another local model on the same clip.
Transcribe the full recording
Import the full copy of the video and use the settings that handled the clip. Export as plain text. Keep a raw copy and make corrections in a separate file so you can compare them later.
Listen before you save the reviewed copy
Replay the opening, a difficult passage, and the final line. Fix names and numbers you can verify. Mark anything you still can’t hear with its time so you can find it again.
- The beginning and end match the recording.
- You checked names and important numbers against the audio.
- The text filename identifies the video cut.
Optional: Export a subtitle file
If you need captions, export SRT and open it with the video. Check one cue near the start and another near the end so you can catch timing drift.
Plan a local transcription from sample to export
Use this with your Mac, video format, and local tool to get a run sheet you can verify on a short excerpt.
I want to transcribe [video name, format, duration, and spoken language] on my Mac using [MacWhisper local mode or whisper.cpp]. My setup is [Mac model or chip, macOS, app or CLI version, and model if known]. I need [plain text and SRT or VTT]. Give me a run sheet from a short sample to the full recording. The sample should include [name, speaker change, or noisy bit] so I can compare the text and timing against the audio. Use a local processing path. If you can’t verify my version’s menus, CLI flags, model compatibility, or export options, tell me what to check in the app or its help before proceeding. If Terminal applies, show which file to convert and how to keep raw and reviewed exports separate. End with checks for names, numbers, unclear speech, and caption sync at the beginning, middle, and end.
Mark transcript lines I should listen to again
Use this at the correction step with a permitted text excerpt and a glossary to make a listening checklist.
Review this text I can share [transcript excerpt with timecodes] using [names, product spellings, and other known terms]. Make a listening list for possible name, number, or meaning errors. Give the exact timestamp, original wording, possible correction, and why it needs a replay. Distinguish a known spelling from a guess at what was said. Preserve the transcript’s timing and mark overlapping or unclear speech for me to listen to. If a timecode is missing, quote a short search phrase I can find. Leave factual corrections pending until I compare them with the recording.
Check my caption file and give me a sync test
Use this with your reviewed text and timed caption export to check cue structure and plan playback checks.
Here are [SRT or VTT text], [reviewed transcript], and [video duration if known]. Flag malformed or out-of-order timestamps, overlaps worth checking, cues beyond the duration, missing lines, and captions that appear too long to read. Show the cue number or timestamp and the smallest edit. Then give me a playback checklist for the opening, a speaker change, and the last line. If the caption text alone can’t tell you the size or direction of an offset, show how I should record that in playback before shifting cues.
Troubleshooting
Why does the CLI reject the input video?
Check the CLI documentation for its supported audio format and convert a working copy before retrying. Verify that the converted audio plays and has the same duration as the original cut.
Why are the captions increasingly late?
Verify the subtitle file matches the exact video cut and audio source. If the cue offset grows, compare durations and transcribe from audio taken directly from the final video.