Transcription
StoryFolder transcribes with Whisper running as a process on your own computer, using audio taken from the file already on your drive. Every line is joined to the shot it was said in, so the transcript sits beside your storyboard instead of next to it.
Transcription is free on every plan.
Does my video get uploaded?
No. The audio never leaves your computer. Speech recognition runs locally and so does speaker identification. No third-party transcription account is involved.
The network is needed once, at the start, to download the recognition engine for your platform and a model file. After that a transcription runs with the network off.
That's a statement about where the bytes went, not a compliance certification. And it's about transcription only — AI Autofill sends frames to StoryFolder's server and on to a hosted vision model. See Local vs cloud.
When does transcription happen?
By itself, after import. Automatic transcription is on out of the box, and a video is queued for transcription as soon as its import reaches Complete.
It's a separate job from the import on purpose: a transcription that fails, is cancelled, or finds no engine installed can't fail your import. You get the shots either way.
What can I change in settings?
Open Preferences (Cmd+, on Mac, Ctrl+, on Windows) and go to the Transcription pane.
- Automatic transcription — the master switch. Off stops new transcriptions; transcripts you already have stay readable.
- Download the model at startup — on by default, so your first import already has something to run against. Turn it off on a metered connection and the download waits until you first transcribe.
- Language — Auto-detect, or one of 100 languages. Worth setting by hand when the audio is quiet or switches language mid-file.
- Translate to English — transcribe non-English audio straight into English. It needs a multilingual model; the English-only ones refuse the task.
- Identify speakers — off by default, and slower when it's on. See Speakers.
- Voice activity detection — on by default. Keeps the recogniser off the stretches with no speech in them, which is what stops timings drifting over long music passages.
- Model — see Transcription models.
Where do I read the transcript?
In the Transcript tab of the workflow bar, or press R. The tabs run Library · Shots · Transcript · Statistics · Export.
There are five view modes:
- By shot — the transcript laid against the shot grid. This is the one that matters.
- By speaker — grouped for screenplay-style reading.
- Continuous — paragraphs, for proofreading.
- Captions — one row per caption, for subtitle QC.
- Plain text — for drafting and pasting.
The Display popover controls speaker names, timestamps, Highlight low confidence, and Follow playback, and holds the Re-transcribe… action.
How do I re-transcribe a video?
Open Display → Re-transcribe… from the transcript toolbar. Re-running replaces the whole transcript, so it asks first and tells you how many hand-edited lines it would discard. If you've edited lines, the first attempt is refused and the button becomes Replace it anyway — so you can only force it after reading why.
The confirmation lets you override the language and the task (transcribe or translate) for this run only, without touching your global preference. An Advanced settings disclosure adds the model, speaker detection and voice activity detection for the same one run. Anything left at its default changes nothing.
What should I check before I trust a transcript?
Names, numbers, and the places where the audio was hard. Whisper degrades on heavy accents, on people talking over each other, on noisy location sound, and on specialist vocabulary. Product names and surnames are where you'll notice first.
StoryFolder's answer is a dotted underline under the words the model was least sure about — turn on Highlight low confidence in the Display popover. There's no percentage anywhere, because Whisper's own confidence isn't calibrated well enough to print as one.
On a long project, check the ends rather than the middle: the first and last minutes, the longest shot, one shot with music and no dialogue, and one where two people overlap.
A video with music and no speech comes back as No speech found — that's a result about the audio, not a failed run.
Which platforms is this available on?
The recognition engine is published for macOS on Apple silicon, macOS on Intel, and Windows x64.