Use Transcribe Only

Prerequisites

• The speech model installed, unless you have RoughCut Pro and choose Cloud transcription for this job. See Install the local speech model.
• A supported video or audio file. See Choose video and audio inputs.
• No AI provider key. Transcribe Only never runs the AI editor, so no Gemini or OpenAI key is needed on the on-device path.
• No purchase. This tool is not gated by the trial. See How the trial and purchases work.

Steps

1. Choose Transcribe Only in the sidebar, under Studio. The screen is headed Transcribe a video or audio file.
2. Add your file. Drop it on the zone that reads Drop a video or audio file here, click Browse…, or send it from Final Cut Pro and pick Transcribe in the chooser sheet. See Add RoughCut as a Final Cut Pro share destination.
3. Set Language. Auto-detect is the default and is right for most clips; choose English or Arabic when detection guesses wrong.
4. Leave Recording environment on Standard. Silence detection still runs on this path, so the threshold genuinely affects how the audio is segmented for transcription.
5. Leave Transcription on On-device unless you have a reason to use the Pro Cloud path. On-device transcribes with Whisper on this Mac and uploads nothing.
6. Click Transcribe.
7. When the job finishes, read the result and export Export SRT… or Export Text….

Expected result

The job runs through audio extraction, silence detection, and transcription, then completes. You get a Your transcript is ready banner with a word count, a duration, the detected language, the full transcript, any flagged time ranges, and two export buttons. See Read and verify a transcript result.

What Transcribe Only skips

This is not a rough cut with the cutting turned off. RoughCut leaves out whole stages:

Because there is no cut plan, a finished Transcribe Only job has no review list, no Recompute Cuts, and no Export for Final Cut Pro…. Its toolbar offers Reveal in Finder and the two transcript exports. Transcribe Only also skips the media probe, so it never raises the variable-frame-rate warning that a cutting job can.

It is free, and it never spends a rough cut

Transcribe Only is the one part of RoughCut that is not metered at all:

• It is never gated by purchase. It runs whether or not you have bought the unlock and whether or not your free cuts are exhausted.
• It never consumes one of the five included rough cuts. Only a completed cutting job does that.
• There is no cap on how many files you transcribe.

The Pro options that appear on this screen are still Pro. Smart under Recording environment and ✦ Cloud (OpenAI) under Transcription open the Pro paywall for non-subscribers, and the selection stays where it was. Everything else on the screen is free.

Where your audio goes

The screen's own subtitle claims transcription here is on-device without exception. That subtitle is wrong, and it is the one line on this screen you should not trust. The same screen renders the Transcription picker, and its ✦ Cloud (OpenAI) option works on Transcribe Only jobs exactly as it does on cutting jobs.

The accurate statement is per-path:

• On-device — the default, and the private path. RoughCut transcribes with Whisper on your Mac. Nothing is uploaded and no API is called.
• ✦ Cloud (OpenAI) — a Pro, per-job choice. When you select it, that job's audio is sent to OpenAI's transcription model under your own OpenAI key, and OpenAI bills you at their per-minute rate. Only audio is sent; no video is ever uploaded.

The choice is made per job and is visible on the screen before you press Transcribe. See Choose on-device or cloud transcription and What data leaves your Mac.

Transcribe the cleanest audio you have

RoughCut says it on the screen itself: if you recorded a separate microphone track, transcribe that instead of the camera file. A lavalier or shotgun track that is closer to the speaker and further from room noise produces a materially better transcript than the same words picked up by a camera mic across the room.

Audio files are first-class input here. wav, mp3, and m4a are all accepted, and an audio-only file sent from Finder or Final Cut Pro goes straight to this screen — the chooser sheet does not offer to make a rough cut from audio, because there is no video to cut.

Common mistakes

• Waiting for the model when you did not need to. Transcribe Only needs the speech model on the on-device path, exactly like a rough cut. If Transcribe is disabled with the hint Download the Whisper model in Settings first., install the model first.
• Assuming Cloud is unavailable here. It is available, it is Pro, and it does send that job's audio to OpenAI. Check the Transcription picker before you start a sensitive job.
• Transcribing the camera file when a mic track exists. The mic track will almost always transcribe better.
• Expecting a timeline. There is no FCPXML on this path. If you want cuts, use Create your first rough cut instead.
• Expecting an editable transcript. RoughCut exports the transcript; it does not offer a full transcript editor in the current version.

Troubleshooting

If the job fails outright, read the message on the job before deleting anything, then see Troubleshooting by symptom.