Cut the ums, the dead air and the take you started twice
Bring a recording. See every cut before it happens. Keep the take you meant.
Already have an account? Sign in
Free to start · No credit card
Cut the ums, the dead air and the take you started twice
Bring a take you already recorded. Kyndrify reads it into words, then lists every um, every long pause and every sentence you started over. You wave back anything you want to keep, and the rest comes out in one move.
The filler list is English. A take in another language still gets its long pauses and its repeated starts found.
Part of The video editor.
How do I get the filler words out of a video I already recorded?
Open the editor, bring your recording, and Kyndrify reads the talk into words. It then lists four things: long pauses, ums and uhs, sentences you started over, and slips it judged as mistakes. Each one is a row you can wave back in. Reading and reviewing are free. Only the finished export costs credits.
Who it is for
Anyone with a recording of themselves talking, on every plan including Free.
What you need
A video file you already have, as mp4, mov, webm, m4v or mkv. Or a video you already rendered here, which is waiting for you when the editor opens.
What it costs
Reading the take into words costs nothing, and so does every pass over the cuts. A finished export costs 20 credits per second of finished video on every plan, with a minimum of 60 credits. The Free plan's one-time 8,400 credits cover about seven minutes of finished video.
What it does not do
The filler list is English, so a take in another language gets its pauses and its repeated starts found and no filler rows. It only takes words out. A line you never said properly is a re-record, not a cut.
How it goes
- 1
Bring the take.
Drag in a file you filmed, or open a video you already rendered here. It does not have to be short and it does not have to be tidy.
- 2
Let it read.
Kyndrify reads the talk into words with the timing of every word. That costs nothing, and it happens once per file, so trimming or reordering never asks for it again.
- 3
Review, then export.
Play any row. Wave back the pause you meant to leave. Apply the rest in one move, and export when it sounds like you.
A minute of talk, before and after
Here is a paragraph of the way people actually talk. The um is a filler standing on its own. The gap after today is dead air, read off the clock rather than off any word list. The thing I wanted to say was started, lost and started again, so it is a repeat, and the first run is the one that goes. And the clause that trails off is a thought that was dropped and never picked back up, which is the kind the second read judges as a mistake.
What you said
- Right, um, the thing I wanted to say, the thing I wanted to say today ... is that we shipped it this morning.
- And the reason that matters, uh, well. It took three weeks.
What comes out
- Right, the thing I wanted to say today is that we shipped it this morning. It took three weeks.
- Nothing re-voiced, nothing re-timed, and nothing added. The words that stayed are the ones you said, in the order you said them, in your own voice.
Filler words, dead air, restarts and slips
Long pauses
Dead air at the front while you walked back to the chair, the gap where you lost your place, the three seconds after a sentence landed. It reads the gaps off the clock, not off a word list, so a pause is a pause in any language.
Filler words
Um, uh, erm and hmm, however long you held the vowel. Uhhhh is the same hesitation with the key held down and it is caught as one. Those four are found at every setting. At the thorough setting it also offers like, plus the padding pairs: you know, sort of, kind of, I mean. The pairs count only when the two words are said together as one unit.
Sentences you started over
The false start, the restatement, the third attempt at an opener. The plain rule is the strict one: the same words said again within a few seconds, with nothing but hesitation in between, and the first run goes. The second read goes further. It will still call it a restart when the two attempts do not match word for word, so you can reorder the middle on the second go and it is still caught, as long as the second go still starts or ends the same way.
Slips it judged as mistakes
A second read goes over the take looking for four shapes: you started over, you corrected yourself, you dropped a thought and never came back to it, or you piled a cluster of fillers into one breath. Each one comes back with the words that would go and a plain sentence saying why.
Which copy stays when you said it twice
When you start a sentence over, both copies hold the same words. Whichever one goes, the finished script reads the same. What differs is how each one sounds.
Two rules answer it, and the rows tell you which one caught what.
The plain repeat rows work the simple way. Same words, said again within a few seconds, with nothing but hesitation in between. The first run goes and the last, fluent delivery stays.
The second read is where the pace rule lives. Here is what actually happens on a take. You start a sentence, you lose it, you go back, and you rush through the bit you already said to get back to where you were. That second copy comes out faster than the way you normally talk, and then it settles once you reach new ground. So when the second read judges a restart, Kyndrify keeps the calm copy of the words you repeated and joins it to the second go at the point where the rushing stops. It reads your own speaking pace off the whole take, weighs the retry's opener against it, and the one that rushes is the one that goes. On a two-word stumble, where the same two words came out twice in a row, the one that drags goes too. When both copies sit at your pace there is nothing to prefer, so the first one goes.
That rule came from listening to a real take and asking which edit sounded like a person.
How careful do you want it to be
There are three settings, and standard is already picked. The ladder moves how long a silence has to be before it counts as dead air, whether the padding words are offered at all, and how far apart two runs of the same sentence can sit and still read as one restart.
Gentle asks for a longer silence before it calls it dead air, leaves the padding words alone, and keeps the tightest window on two runs of the same sentence.
Standard uses the same idea of dead air the one-tap tidy uses, so the two can never quietly disagree with each other, and gives a restart a little more room.
Thorough takes shorter gaps, adds the padding words, the real ones that are usually filler and occasionally meaning, and gives a restart the widest window, so a sentence you came back to after a longer think still reads as one. That setting exists because you get to look first. Offering the padding words at all would be rude without the review.
The ums, the repeats and the judged slips are found at every setting. The switches beside the ladder are what turn those off: one for repeats and one for the judged slips, on their own, because some people want the pauses gone and the personality kept.
Where a cleaned take goes next
Once the stumbles are out, the take can carry on. Kae, the AI creative director plans the script, the shots and the voice before anything renders. Talking-head video lets you say the script once and let your Digital Twin film it.
The voice room holds narration, cloned voices and the transcript tool for when you want the words as a file. Creators who post every week shows what a week of content looks like without a filming day.
What it will not do
The filler list is English.
A take in another language still gets its long pauses and its repeated starts found: a pause is read off the clock, and a repeat is the same words coming back, whichever language they are in. Neither needs the list. The filler rows do need it, so it gets none.
It takes words out. It cannot put words in.
If a sentence never came out right on the day, there is no good copy to keep and no cut will save it. That one is a re-record, and we would rather say so than sell you a fix.
It does not judge your content.
It finds hesitation, dead air, restarts and slips. It has no opinion about whether the point you made was a good one.
A free export carries a small visible mark.
The reading and the reviewing are free on every plan, so the cleanup itself is never the thing behind the paywall. The clean picture is.
Questions people ask before they upload
- Does it cut things automatically, or do I get to look first?
- You look first. Every group is a row with the words printed on it, and you can play any of them before you decide. The one exception is honest and visible: when you use the one-tap tidy, the slips it was sure about come out with the quiet moments, and they are listed at the top of the window under Already tightened so you can put any of them back. Kae plans on the storyboard; this room is manual, and you press every button in it.
- What counts as a filler word?
- Um, uh, erm and hmm, with any length of vowel, as whole words. Those four are offered at every setting. Like sits in a second set with the padding pairs, you know, sort of, kind of and I mean, and only the thorough setting offers that set, because those are real words that sometimes carry meaning. The pairs count only when the two words are said together as one unit. You look at all of it before anything moves, which is what makes offering the second set honest at all.
- My take is not in English. Does any of this work?
- Some of it. A long pause is read off the clock, and a repeated start is the same words coming back, whichever language they are in, so both of those still work. The filler rows do not, because that list is English. We would rather say that here than have you find out after signing up.
- It cut something I wanted to keep. Can I get it back?
- Yes. Anything still waiting is a row you simply do not apply. Anything the tidy pass already took has a Put back control beside it, and putting one back re-plans the rest around it.
- I said the same sentence twice. Which one does it keep?
- That depends on which rule caught it, and the row tells you. A plain repeat keeps the last delivery, the fluent one, and drops the earlier runs. When the second read judges it a restart, it keeps the calm copy of the words you repeated and joins it to the second go at the point where the rushing stops, because the rushed copy is the one that sounds wrong when you play it back. On a two-word stumble, the copy that drags goes as well. If both copies sit at your pace, the first one goes.
- What does this cost?
- Opening the editor, reading the take into words and every pass over the cuts cost nothing, on every plan including Free. Opening the room does not use up one of your Kae planning sessions either: that only counts once you actually ask Kae to plan something, and cutting a take you already filmed never does. A finished export costs 20 credits per second of finished video on every plan, with a minimum of 60 credits. The Free plan's one-time 8,400 credits cover about seven minutes of finished video.
The tools that do this
More on this from the blog
Your take is probably fine. You are just in it four times.
Bring the recording, look at what it found, and keep the parts you meant.