The podcast recording mistakes no editor can fix

An editor's job is subtraction and rearrangement. We take what was recorded, remove the parts that do not belong, move the parts that do, and balance the result so it sounds and looks as though it was always meant to be that way. Nothing in that description creates information. That is the whole idea behind this article, and it is the single thing that separates a recording fault you can stop worrying about from one that will cost you the episode.

The faults that reach our edit suite, across a library of shows that runs from a handful of episodes to 79 on one title, fall neatly on either side of that line. On one side sit the problems an edit can absorb, which is most of them. On the other sit the problems where the thing we would need in order to fix them was never captured, and no amount of skill, software or budget puts it back. This guide is about the second kind: what they look like in the raw files, why they are permanent, and the short physical check that prevents every one of them.

Which recording faults can an editor actually fix?

Almost everything that happens during a take. A cough in the middle of a sentence, a stumble that needs a second run, a dog in the next room, a phone buzzing on the desk, a guest who talks over the end of a question, a host who says "um" every eleven seconds: all of these are edit problems, and they are solved by cutting, trimming and rearranging material that exists. A take that went badly is not a fault at all if there is a second take after it. Uneven levels, where one speaker is louder than the other, come up in nearly every multi-mic session and are handled with gain and compression before the first cut is made. A room that sounds a bit boxy can be tamed. A camera set to the wrong white balance can be corrected, and a slightly crooked horizon can be straightened.

What all of those have in common is that the signal is still there. It is degraded, or it is surrounded by things that need removing, but the words the person said and the face they said them with are intact in the file. Editing works on what exists. Give an editor a rough take of a good conversation and you will get a good episode back. That is the routine, and it is why the video and audio editing service exists at all: the recording does not have to be perfect, it has to be complete.

What is the difference between a problem and a missing thing?

A problem has a signal underneath it. A missing thing has nothing underneath it. The test is simple: could a patient person with the raw files and unlimited time eventually get to the right result? If yes, it is a problem, and it costs hours. If no, it is missing, and it costs the episode, or that part of it.

The clearest example we have seen is a guest track that never arrived. On one four-episode show, three of the four episodes came to us with the guest's separate track absent from the finished multicam transcript. The transcript showed long blank stretches with timestamps and no words, a range like 00;06;44;13 to 00;07;06;23 with nothing in it, while the host's side of the conversation made it obvious that someone was answering. Only one of the four episodes had a clean two-way transcript.

That show was lucky. The guest's audio did exist, as separate per-speaker raw files that had been recorded locally, and the fault sat in what got assembled into the multicam rather than in the recording session itself. So the missing thing turned out to be a problem after all, and a route back existed: rebuild the guest's side from the raw files. But the reason it was recoverable is exactly the point of this article. Someone had made sure each speaker was captured separately, to a file, before the session ended. Remove that one decision and the same three episodes would have had twenty-second holes wherever the guest spoke.

Why do clipped or crossed mics never come back?

Because the recorder wrote down the wrong thing, and it wrote it down permanently. When a microphone's gain is set too high and a speaker laughs or leans in, the loudest peaks exceed what the recorder can represent, so it writes the maximum value instead. The top of every wave is flattened. What arrives in the edit is a buzzing, papery distortion on every emphasised word, and no software can guess what the shape of that peak was supposed to be. It can smooth the edges. It cannot restore the sound.

Crossed mics do a subtler version of the same damage. Two people sharing a mic, two mics pointed at each other across a small table, or a mic placed so far from the mouth that it hears the room more than the person: all of these produce a track where the voice is mixed with things that cannot be separated from it. Bleed from the other speaker, the echo of the walls, the rumble of a desk. Noise reduction can remove a constant hum. It cannot remove another voice sitting on top of the one you want.

We saw the outcome of this in one episode with two interviewees. The first person's segment transcribed cleanly and gave the show five complete, usable quotes for its clips and notes. The second person, in the same room, in the same session, came back as single disconnected words: "Right." "Yeah." "Was also." A human listening to the raw audio could make out no more than the transcript did. There was no edit to be done on that segment, because there was nothing to edit toward. The only routes forward were to re-record that person or to go back to the original synced audio and see whether a different capture held more.

Mic placement and gain are the two decisions that put you on the right side of this, and both are made before anyone says a word. Our microphone guide covers the settings in detail. The short form: the mic close to the mouth, the gain set with headroom to spare while the speaker talks at real conversational volume, and someone wearing headphones listening to what the recorder hears rather than what the room sounds like.

What happens when nobody is in the frame?

The picture is part of the product now, not a bonus layer. Ofcom's 2026 audio report found that 68% of weekly podcast listeners in the UK listen to podcasts that include video, so for most of the audience the video version is part of what they signed up for, whether or not they watch every minute of it. A video fault is therefore an episode fault.

The framing faults an edit cannot fix are all versions of the same thing: the pixels that should have held a face hold something else. A guest who sat to the left of where the camera was pointed, so their shoulder is in shot and their face is not. A head cropped at the eyebrows because the tripod was set for someone shorter. A camera left on autofocus that spent the interview hunting between the face and the bookshelf behind it. A lens with a thumbprint on it. A wide shot that does not include the person once they lean back. In every case there is no second angle to cut to, because the second angle is the one that failed.

An editor can crop in, which costs resolution. An editor can stay on the other camera, which turns a two-camera conversation into a lopsided one where the guest is only ever heard. What an editor cannot do is put the person back in the picture. Check every participant's frame before the conversation starts, with the person sitting how they will actually sit, and again after any break where someone stood up.

Where does "we'll download it later" go wrong?

Later is where files disappear. The most common form of the fault is a recording that only ever existed inside a browser tab: a remote session where nobody confirmed that each participant's machine was writing a local file, the tab was closed, and what was left was a compressed stream with dropouts wherever the connection wobbled. The second most common form is a session where local recording was on, the files were uploaded, and nobody kept the originals, so when the upload turned out to be incomplete there was nothing to go back to.

The reason people skip the local copy is that it is genuinely large. This is what the files actually weigh, taken from our own archive. Raw camera files run about 8.5GB per camera. Synced multicam takes on a three-angle shoot run 9GB to 14GB per angle. Finished masters run from 2.1GB to 16GB depending on length and format, and nearly every finished export is well over 400MB. A single three-camera session is tens of gigabytes before anyone has made a cut. "Keep a local copy" means finding tens of gigabytes of space per session and remembering to do it every time, which is exactly why it gets skipped.

The cost of skipping it looks like this. One show's most recent episode arrived with no transcript, no finished cut and only two raw single-speaker recordings of about 600MB each. That is at least a starting point, and only because those two files were saved. Another show has no watchable finished episode anywhere in its archive at all, only raw synced takes. Whatever that show once published, its archive does not hold it; what it holds is source material.

Local recording, for every participant, to a location you have already checked has room, is the one decision that protects everything else on this list. If one mic clipped, the other person's local file may still hold a clean version of the exchange. If a camera failed, the local recording from the other angle still holds the conversation. Remove the local file and every other fault becomes permanent.

Which admin mistakes cost an episode without touching a cable?

Two, and they are both about names. The first is the file that is labelled as one thing and contains another. We have seen a transcript filed under episode three that documents episode four's conversation, and a set of files named for episode fifteen sitting inside the folder for episode fourteen. Neither fault is technical. Both cost real time to untangle, and both carry the same risk: the wrong episode goes out under the wrong title, or a clip is cut from a conversation that has not been published yet.

The second is the name that was never written down. One episode's spoken introduction has the host describing the guest's company as "Denby Joinery". Every source file for that episode says "Denholm Joinery". An editor cannot tell which is right. We can hear the host and we can read the filename, and they disagree. The lower third, the show notes, the clip captions and the YouTube description all need one spelling, and the only person who knows it is the guest, who is no longer in the room.

Get the spelling of every guest's name, company and title in writing before you record, and put it somewhere the editor will find it. Name the files with the episode number and the speaker before they leave the recording machine. It takes a minute, and it is the difference between an editor working and an editor guessing.

Fault by fault: can the edit fix it?

What is the five-minute pre-record check?

Run this before every record, in this order, out loud if it helps. It is physical rather than administrative: every step involves doing something with your hands or your ears, because meters and status lights are exactly what fooled the sessions described above.

1. Headphones on, and actually listen. Plug wired headphones into the recorder or interface, not the laptop. Listen to what the recorder hears. If you can hear the room more than the voice, move the mic closer. If you can hear the other person in your mic, angle the two mics away from each other.
2. Record a 20-second test and play it back. Do not trust the meters. Have every participant talk for twenty seconds at the volume they will really use, including a laugh. Stop. Play it back through the same headphones. Listen for the buzz of clipping on the loud moments and for the other voice bleeding in on the quiet ones. Adjust, then test again if you touched anything.
3. Check every participant's frame. Look at each camera's picture with the person sitting as they will actually sit, not perched forward for the check. Head fully in shot with space above it, focus locked on the face, lens clean, nothing hunting. If a participant is remote, ask them to show you their frame on the call.
4. Confirm local recording is on for every participant. Not "the session is recording". Each person's machine writing its own file to its own disk. Ask each remote guest to say the word "local" once they can see it is on. If your platform has a setting for it, look at the setting, not the reassurance.
5. Confirm where the files will land. Name the folder before you start: show, episode number, date. Check the drive has room for the session (assume tens of gigabytes for a multi-camera shoot). Agree who is downloading the remote files and when, and do not close any tab until the local files are on a drive you can see.
6. Get the names in writing. Every guest's name, company and title, spelled by the guest, in the run sheet or the chat before you roll.

That is six steps rather than five, and the sixth takes fifteen seconds. Do the whole thing once and it becomes the routine. Skip it once and you will find out which step you needed.

Where does the editor come in?

At everything on the left-hand side of that table. Once the recording is complete, the rest of the chain, from the multicam cut to the clips and the packaging, is work that can be handed over entirely. The complete production guide walks that chain stage by stage. Our editing service takes the raw files you protected with the check above and returns the finished episode, with revisions until you are happy and every file yours to keep. Prices are live on the configurator, or you can book a 15-minute call and tell us about the show.

Bring us a rough recording and we will make it a good episode. Bring us a complete one and we will make it a great one. The five minutes before you press record are the only part of the process we cannot do for you.