A host sent me an episode last year where the guest had recorded into a laptop mic, in a room with a ceiling fan, sitting two feet from an open window. The camera was set to autofocus and hunted every time he leaned forward. The host's own audio was fine. Ninety minutes of great conversation, and half of it sounded like it was recorded inside a washing machine.
We shipped it. It went out on schedule. It is not the best sounding episode that show has ever published, but it is listenable, and nobody in the comments mentioned the audio. That is the whole job when a recording goes wrong: get the episode to a place where the content is what people notice.
I want to write down how we actually approach this, because the advice online is mostly plugin lists. Plugins are the last five percent. The first ninety five percent is decisions.
The first thing I do is decide whether it is worth saving
Before anyone touches a fader, someone on my team listens to the whole thing and answers one question: is the content good enough to justify the work?
This sounds cold. It is actually the kindest question you can ask. A mediocre conversation with broken audio is not worth eight hours of restoration. A genuinely great conversation with broken audio is worth almost any amount of work, because that episode will get shared, clipped, and quoted for a year.
Bad audio on a boring episode is a scheduling problem, not an editing problem. Push it, re-record it, or cut it to twenty minutes of the good part and publish it as a short one. Bad audio on a great episode is where the effort goes.
We can only make that call because human editors watch the whole episode before deciding anything. There is no automated pass that silence-detects its way through a file and hands us a timeline. That is deliberate. A machine cannot tell you that minutes 34 to 51 are the reason this episode exists.
Diagnose before you process
Most rescues go badly because someone starts fixing before they know what is broken. Noise reduction on a clipped file makes it worse. EQ on a room resonance problem makes it thinner without making it cleaner.
Here is the order we work in. It is boring and it works.
- Is it clipped? Look at the waveform. Flat tops mean the information is gone. You cannot restore it, you can only soften how ugly it is.
- Is it noise or is it room? Steady hiss, fan hum, and traffic are noise. Boxy, echoey, hollow sound is room. They need completely different treatment and the fixes fight each other.
- Is it a codec problem? Warbly, underwater artifacts usually mean the recording went through a compressed call and no local file exists. Ask before assuming.
- Is it just level? A surprising number of "bad audio" files are simply recorded at negative 40 dB with a perfectly clean signal underneath.
- Is it sync? If the picture and the sound drift, fix that first. Everything else is wasted work on a timeline that will get rebuilt.
That last one has its own rabbit hole, and I have written up what actually causes desync and how we chase it down separately, because it is the single most common technical mess we get handed.
Restoration is subtraction. Every tool you add takes something away from the signal. The skill is knowing how much ugliness to accept in exchange for how much clarity.
Fix the worst thing first, then stop
The trap in audio repair is that each fix reveals the next problem. Remove the hiss and now you hear the room. Fix the room and now the voice sounds processed. Fix that and the whole thing sounds like a robot reading a phone book.
We work to a target, not to perfection. The target is: a listener with earbuds on a train can follow the conversation without leaning in.
Practically that means gentle noise reduction rather than aggressive. High pass to cut the rumble, but not so high the voice goes thin. A de-esser only if the sibilance is actually a problem. Then a compressor doing modest work, and a limiter to catch the peaks. If one channel is clean and one is wrecked, we treat them separately and match them loosely, not exactly. Matching them exactly usually means dragging the good channel down to meet the bad one.
Loudness normalisation goes last, to a broadcast target, on the finished mix. Doing it early means redoing it.
Cut around the damage instead of processing through it
This is the part most people skip, and it is the part that saves the most episodes.
If thirty seconds of the guest's audio is unusable because a truck went past, do not spend an hour cleaning thirty seconds. Cut it. If the guest coughs, restarts a sentence, and the mic pops on the restart, cut to the host reacting and come back on the clean take.
On a multi-camera show this is much easier, because you always have somewhere to go. That is one of the quiet arguments for shooting more than one angle: angles are not just visual variety, they are cover. You can hide an audio edit under a cutaway and nobody will ever hear the seam.
On a single camera show you have fewer options, so we lean on B roll, a title card, or a short zoom in the frame to disguise the join. It works more often than you would expect. The eye forgives what the ear would have caught.
Video problems are usually easier than audio problems
People panic more about video, but video is more forgiving.
Soft focus is soft focus. You cannot sharpen your way out of it, but you can crop in slightly, add a little grain, and lower contrast so the softness reads as a look rather than a mistake. Bad white balance is a colour grade away from fine. Underexposed footage cleans up reasonably well if it was shot in a decent codec, and if it was shot at 4K there is real room to punch in and reframe.
That is the practical case for higher resolution capture, and it is a different argument than the one people usually make. I went into why in my honest take on whether 4K is worth it for podcasters, and the short version is that the benefit is mostly in the rescue and reframe, not in the delivered picture.
The exception is motion. Rolling shutter wobble, dropped frames, and a camera that overheated and stopped are not fixable. If the camera stopped, you have an audio only section and you should design for it: full screen show art, a static wide, or a clean title card with the audio running under it.
Remote recordings fail in specific, predictable ways
Almost every disaster file I see comes from a remote session. Local recording did not start. One participant used a browser that did not save the local track. The uploads failed and the platform kept only the compressed call audio.
None of this is exotic. It is the same five failures over and over, and every one of them is preventable with a five minute pre call routine. I put the whole checklist in how to make a remote episode not sound remote, and if you record with guests you have not worked with before, that routine will save you more time than any plugin chain.
The single highest value habit: record a backup on a phone in the room, for both ends. A phone voice memo on a table is not a great microphone. It is an enormously better microphone than nothing, and I have used one to rescue a full episode more than once.
Transcripts, chapters, and the clean up nobody thinks about
A damaged recording also damages everything downstream. Automatic transcription falls apart on noisy audio, and the show notes generated from that transcript will be wrong in ways that look confident. I have a longer argument about why I do not hand transcripts straight to a tool and trust the output, and rescued episodes are exactly where that goes wrong fastest.
So on a rescue we tend to do the show notes and chapters pass by hand rather than by machine. Names get misheard. Technical terms get mangled. If the episode is going to be found by search at all, the words attached to it need to be right, which is really a discovery problem more than an editing one.
Vertical clips from a rescued episode need care too. A clip is thirty seconds of your worst audio played at full attention on a phone speaker. Pick clips from the clean stretches. If the best moment lives in the damaged section, sometimes the honest answer is that it does not become a clip.
What this costs, and when to hand it over
I will be straight about the economics. A rescue takes longer than a clean edit. It does not change our price. Starter is $199 an episode for a single camera 1080p edit with one revision, and Studio is $299 an episode for two to four cameras, 4K, and unlimited revisions, up to sixty minutes as the base. First cut lands in 48 hours, revisions come back the next working day, and rush 24 hour delivery is $75 if the calendar has already moved past you.
Whether you hand a mess like this to a studio or to a solo editor is a real decision with real tradeoffs, and I laid out both sides in the agency versus freelancer comparison. Neither answer is wrong. The wrong answer is spending your Sunday learning spectral repair on the episode you needed to publish Friday.
If the rescued episode is going to YouTube, do not let a great save die on a weak thumbnail. My sister studio ClickTheory handles that side, eight thumbnails for $250, and on a show that is trying to grow the packaging matters at least as much as the polish.
The honest closing thought
You will not save every episode. Some files are just gone, and the mature move is to say so, apologise to the guest, and book the re record.
But most bad recordings are not gone. They are messy, and messy is a workflow problem. Listen first, decide if it is worth it, fix the worst thing, cut around the rest, and stop before you process the life out of it.
If you have an episode sitting in a folder because you do not want to open it, send it over and I will tell you honestly whether it is worth saving. samy@pixelsamy.com. I am in Dubai and work with shows anywhere.