Transcribe Audio and Video Files

Drop a File, Get the Text

Dictation isn’t the only way speech reaches your Mac. Interviews, voice memos, lectures and screen recordings are already sitting in files. VocaMac transcribes them with the same speech model you dictate with.

1. Click the VocaMac icon in the menu bar and choose Transcribe File.
2. Drop an audio or video file onto the Transcribe a File window, or click Choose File….
3. Transcription starts right away. When it finishes, Copy the transcript or Save… it as a plain-text file.

Any audio or movie format macOS can open is accepted, up to 500 MB and 30 minutes per file. The window tells you where the work happens: “Processed on this Mac. Uses your selected speech model.”

What Applies to Files

File transcription uses your selected model and language, translation to English if you turned it on, and your custom vocabulary. It does not run transcript clean-up, snippets or formatting, so you get what the model heard. Transcribed files count towards your usage stats but are not added to dictation history.

For the fastest results on long English recordings, try a Parakeet model; for other languages, see Speech Engines for which model fits.

From the Command Line

VocaMac’s app bundle doubles as a headless command-line tool. It never opens a window, never touches the microphone, and never downloads a model, so it suits scripts and batch jobs:

/Applications/VocaMac.app/Contents/MacOS/VocaMac --transcribe-file interview.m4a --json

The JSON response includes text, model, engine, detected_language, duration_seconds and audio_length_seconds. Other useful forms:

# pick a model and a language
/Applications/VocaMac.app/Contents/MacOS/VocaMac --transcribe-file memo.wav --json --model small --language de

# see which models are installed and which is selected
/Applications/VocaMac.app/Contents/MacOS/VocaMac --list-models --json

--json is required. Leave out --model and --language to follow the app’s current settings. The model must already be downloaded in the app. Errors are written to stderr as JSON with an exit code: 2 for invalid arguments, 3 for audio that can’t be read, 4 for a missing or unsupported model, and 5 if transcription fails. The same 500 MB and 30-minute limits apply.

Privacy

With an on-device model, which is the default, the file is transcribed on your Mac and nothing is uploaded. If you selected Custom Endpoint as your speech model, the file’s audio is sent to the server you configured, just like a dictation would be. Point Custom Endpoint at a server on this Mac, on a trusted local network, over Tailscale, or over HTTPS.