Podcast Recording and Editing: The Editor Is the Cheap Part

People building a first podcast tend to budget in the wrong order. They compare editing software, because editing is the part that looks like work, and they record the interview in whatever call app was already open. Then they discover that no amount of editing recovers a voice that arrived as a compressed stream with the room gated out between sentences.

The editor is the cheapest component in the chain and the least able to fix anything upstream of it. That is the whole argument, and it is worth doing the arithmetic before buying either.

What a remote call does to a voice

A standard video call is engineered for intelligibility at low bandwidth, not for fidelity. It compresses hard, it applies automatic gain, and it runs a noise gate that mutes the channel when the guest is not speaking. The gate is the destructive part: it does not just remove the room tone, it removes the breath before a sentence and the tail after it, so every cut in the edit lands on silence instead of on a natural decay.

The result is usable for a meeting and painful for an hour of listening. Listeners describe it as "sounds like a phone call," which is accurate, and they stop listening.

There is also the drop-out problem. Packet loss during a call is a moment of garbled audio in the recording, and there is no second copy to cut to. The interview is simply damaged at that point, and the only fix is to not use that section.

Four ways to capture, and what each costs

The cloud studio option exists to remove the hardest part of the double-ender: telling a non-technical guest to install software, pick the right input, and hit record. A tool like Riverside records locally in the browser and uploads afterward, which gets you most of the quality of a double-ender with one link instead of a setup call. The trade is that you are trusting an upload at the end of the session, and the failure is quiet — you finish, everyone says goodbye, and one track never arrives.

Whichever you choose, the rule is the same: each person's audio must be captured on their own machine. Anything that travels over the network in real time is already lossy by the time it reaches you.

The arithmetic of one re-record

Fifty-five minutes with a guest who has a job. Suppose the recording turns out unusable — the guest's track is a phone-quality stream with two dropouts in the most important answer.

Rescheduling costs a week of calendar time across two timezones, plus the episode slipping a slot. The session itself costs two people two hours each, counting setup and the buffer around it. At an illustrative sixty dollars an hour of equivalent time, that is two hundred and forty dollars, before counting the guest's goodwill, which is the part you cannot rebook.

A capture tool that records locally costs less per year than that single re-record. This is not a close comparison, and it is the reason the budget should sit in capture rather than in the editor, where the monthly fee feels more like a real decision.

The file-size arithmetic points the same way. Forty-eight kilohertz, twenty-four bit, mono is about 144 kilobytes per second, so roughly 8.6 megabytes a minute and just over 500 megabytes for an hour. Two tracks for a one-hour episode is about a gigabyte before editing. Storage at that scale is not a constraint, which means there is no reason to record anything less than WAV.

Why the editor cannot save you

Editing software removes things. It removes silence, mistakes, ums, and the two minutes where the guest repeated themselves. It cannot add information that was never captured.

Noise reduction is the clearest example. It works by subtracting a noise profile from the signal, and it takes some of the voice with it — usually the breath and the top of the sibilance, which is exactly what makes a voice sound present. De-reverb is worse: it makes a room sound like a towel. Applied heavily, both produce a recording that is quiet and dead, and listeners read that as "bad microphone" when the microphone was fine.

The same caution applies to the "studio sound" processing that transcript editors now offer. It is genuinely impressive on a mediocre recording and unnecessary on a good one. Use it at a fraction of its default strength, or not at all, because it flattens the differences between speakers and makes every episode sound like the same show.

Transcript editing versus a waveform

Descript's pitch is that you edit audio by deleting text, which collapses the first pass of an edit from hours to minutes. For a solo host with a two-person interview, that is a real reduction: cutting a false start becomes selecting a sentence rather than hunting for a waveform edge.

What it does not remove is the second pass. Levels between two speakers still need matching, music still needs ducking, and the loudness target still needs hitting. Those are the same regardless of the tool.

Audacity is free, honest, and unchanged in the ways that matter. Its noise reduction is destructive and its multitrack workflow is clumsy, but for a single-track interview with light cleanup it is entirely sufficient, and the price means you can spend the money on a better microphone instead.

The order of returns is microphone, room, capture, then software. A hundred-dollar dynamic microphone in a carpeted room will beat a studio condenser in a kitchen, and neither matters if the recording arrives compressed.

The minimum rig, and when to buy it

One dynamic microphone per person on a boom arm, close to the mouth. Headphones, so the guest does not hear themselves through speakers and create an echo. A room with soft surfaces. Local recording at 48 kHz, 24-bit, one file per person, and a clap or a countdown at the start so the tracks can be aligned.

That is the whole setup, and it costs less than the first year of most podcast hosting.

The buying rule is to ship five episodes before spending anything beyond a microphone. Most podcasts fail at episode four, and the failure is almost never audio quality — it is that the format was not interesting enough to sustain. Software will not tell you that, but five episodes will.

Rule of thumb: record locally in WAV, fix the room before you fix the chain, and if you have to cut a line from the budget, cut the editor subscription before the microphone. The editor is where you spend time. Capture is where you spend money, because it is the only part you cannot redo.