Cutting and cleaning
A recorded conversation has silences, hesitations, repetitions and the odd cough. Removing what's not needed makes the episode tighter, but if you cut too much it sounds robotic. Here you'll learn what to cut, what to keep and how to make cuts nobody notices.
What to cut and what not to
The goal of editing isn't that nobody ever hesitates or breathes, but that the listener doesn't lose the thread or their patience. A podcast is a conversation, and real conversations have pauses, laughter and half-finished sentences. Remove what gets in the way, not what makes it sound human. Try it in the simulator above: cut sections and see how the length changes and whether the result still sounds natural.
| Element | Cut it? | How |
|---|---|---|
| Long silences (more than 1–2 s for no reason) | Yes, shorten them | Leave a few tenths of a second, not zero |
| Filler words (“um”, “uh”, “like”) | The excessive ones | Remove the ones that pile up; the odd one is natural |
| Repetitions and sentences started twice | Yes | Keep the good, complete version |
| Coughs, throat-clearing, sneezes | Yes | Cut and join, or replace with room tone |
| Breaths | Only very loud ones | Turn them down instead of deleting them |
| Laughter and reactions | Almost never | They add life; trim only if they drag on |
| Tangents that add nothing | Depends on the format | Decide with the listener in mind, not the guest |
Silences, filler words and repetitions
Long silences are the first thing you notice in the waveform: flat zones between blocks of speech. Shorten them, but don't remove them completely: a half-second pause before an important answer gives weight to what's coming. If you butt every sentence together with no air, the listener gets worn out.
With filler words, the criterion is density. An “um” now and then bothers nobody; five in one sentence do. Remove the ones that get in the way of understanding and the ones at the start of answers, where they stand out most. As for repetitions, when someone starts a sentence, corrects themselves and says it again, keep the second version if it's complete.
Cutting is for cleaning up, not for changing the meaning. Never splice parts of different sentences so a guest says something they didn't say, and don't remove a nuance that changes their opinion. It's a matter of ethics and trust.
Coughs, noises and breaths
A cough or a bump on the desk gets cut and joined. If the cut leaves an odd gap, fill it with a few seconds of room tone, the room's ambience recorded in silence, so the background doesn't suddenly vanish.
Breaths need care. If you remove them all, the voice sounds unnatural, as if the person didn't need air. The usual approach is to leave them and, if one is very loud, turn it down by about 6–10 dB instead of deleting it. Some programs have effects that detect and soften breaths automatically; use them sparingly and check the result.
Don't
Delete every breath: sentences run into each other with no air and the voice sounds glued together.
Do
Keep the breaths and turn down only the loudest: it sounds natural and clean.
How to make a cut nobody notices
A badly made cut sounds like a little “click”. It happens because the wave jumps suddenly from one value to another. There are two ways to avoid it:
- Cut at the zero crossing: the point where the wave crosses the center line, with no value. Many editors have an option to snap cuts to that point automatically.
- Short fades: a fade or crossfade of between
5and20 msat each join smooths the jump. It's the most reliable method when there are several tracks.
Also, cut in the pauses, not in the middle of a word, and pay attention to intonation: if you join the end of a rising sentence to the start of another, the ear notices something doesn't fit. Listen to each cut a second or two before and after.
If each person has their own track, you can cut just one of them. That way you remove the guest's cough while the host is talking, or overlap the turns slightly so the conversation flows. It's one of the big advantages of recording on separate tracks.
Editing with the transcript
More and more programs transcribe the audio automatically and let you edit by deleting text: you strike out a sentence in the transcript and it disappears from the audio. They can also detect and remove filler words and silences in one click. For a first pass on a long interview, it's a huge help.
But it has limits. The transcript doesn't hear breaths, changes in tone or noises that aren't words, and automatic filler-word removal can leave abrupt cuts or swallow syllables. Use the text for the rough cut and always finish by listening to the cuts in the waveform.
Keeping a natural rhythm
A good cut is one nobody notices. To get there, edit in two passes: a quick first one to remove the obvious (mistakes, repetitions, long silences) and a second one listening straight through the episode like a listener, without looking at the screen. In that listen you'll notice odd joins and sections that drag.
As a reference, a well-cleaned conversation usually loses between 5 and 15% of its original length. If you're cutting a lot more, check that you're not also removing the speaker's personality.
- Cut what gets in the listener's way, not what makes the conversation sound human.
- Shorten long silences, but leave air between sentences.
- Turn down loud breaths instead of deleting them.
- Cut at the zero crossing or with 5–20 ms fades, and always in pauses.
- Text-based editing helps with the rough cut; finish by listening.
Your turn: five cleaner minutes
Take five minutes of a recorded conversation and clean them up without any cut being noticeable.
- Note the starting length and remove long silences, repetitions and coughs.
- Turn down two or three loud breaths instead of deleting them, and add short fades at each join.
- Listen to the result without looking at the screen and note where you notice a cut; fix it and compare the final length.
Frequently asked questions
Should you remove breaths from a podcast?
Not all of them. If they're removed completely, the voice sounds unnatural; the usual approach is to keep them and turn down only the loudest.
How do I remove “ums” and filler words from audio?
You can cut them by hand in the waveform or use an editor with a transcript that detects them. Remove the ones that pile up and check each cut by listening to it.
Why do I hear a click when I cut audio?
Because the wave jumps suddenly at the cut point. You avoid it by cutting at the zero crossing or applying a very short fade at the join.