Cleaning dialogue
Find and cut bad takes
You start a sentence, fumble it, stop, and say it again. On camera that is two takes of one line, and the first is dead weight. This tool reads the transcript, finds the pairs, and offers to drop the one you abandoned.
How it decides
It compares what was said, not where it sits. Two nearby stretches of speech are a retake when enough of their words line up in the same order, allowing for the fact that a second attempt is rarely word-for-word: you drop a word, add one, change the ending.
At least four matched words are required, and on a short line that floor drops to the length of the shorter attempt, so “so today we’re — today we’re looking at” still registers.
Step by step
- Open Remove bad takes, then set the timeline area and voice track under Choose the source.
- Analyze. Like the other transcript tools, the first run downloads the speech model.
- Each result shows both attempts side by side with the words that matched. Read them; only you know which one you meant.
- By default the later attempt is kept, because a retake is normally an improvement on the one before it. Switch it if the first was better.
- Apply. The discarded attempt is removed from picture and dialogue together.
When a retake is missed
Three common reasons, in the order worth checking:
- Too long between attempts. There is a limit on the gap, because two similar sentences ten minutes apart are usually a deliberate callback rather than a stumble.
- The second attempt rephrased. “Let me start over” followed by an entirely different sentence shares no words with the first, so nothing links them.
- The line was too short. Three words repeated is often just emphasis.
A missed take is not a failure state you have to work around: cut it by hand and carry on. The tool errs towards missing one rather than proposing to delete something you meant to keep.
What it will not do
It will not judge performance. A take you delivered badly but said correctly looks identical to a good one in a transcript. This finds repetition, not quality, and it will never remove a line that was only said once.
What it needs
The local speech model, like filler words and captions. Available on every paid plan, and the transcription stays on your machine.