Podcast and Spoken-Word Audio

How to clean up background noise that changes through an episode in REAPER

By StudioRescue Editorial

Published October 10, 2026

This is the problem that shows up after you have already learned noise reduction. You find a quiet gap, you let ReaFIR build a profile, the first minute sounds clean, and then twenty minutes in the hiss is back, or a passage that was fine now sounds thin and watery. Nothing went wrong with your settings. The noise changed and the profile did not.

Steady noise and changing noise are two different jobs. If yours is genuinely constant, the simpler method is the right one and it is covered in how to reduce steady background noise in a podcast in REAPER. This page is for the other case.

First, prove the noise actually changes

It is worth ten seconds of checking, because the fix below is more work than it is worth if the noise was steady all along and the real problem was something else.

Why one profile stops working

Subtract mode in ReaFIR works by capturing a picture of the noise across the frequency spectrum and then taking that picture away from everything that follows. The Cockos effects guide describes the method plainly: you turn on Automatically build noise profile while a noise-only section plays, then uncheck it once you are satisfied so the profile stays fixed.

Fixed is the operative word. The subtraction is only ever as accurate as the match between the stored profile and the noise actually present.

• If the noise gets louder than the profile, some of it survives. You hear the original hiss coming back through, usually in the gaps first.
• If the noise gets quieter than the profile, the plugin keeps subtracting the full amount anyway, and the extra comes out of the voice. That is the watery, flangey, underwater sound people blame on the plugin being bad.
• If the noise changes shape, say a steady hiss gains a low drone when the compressor on the air conditioner kicks in, the profile is wrong in both directions at once.

REAPER has no documented way to rebuild that profile automatically as the audio plays. This question has been asked on the Cockos forum and the answers from experienced users there have been consistent for years: no one knows a way to do it, and the practical suggestion is to process each section separately instead. That is what the rest of this page does.

The fix: split the episode into noise zones

The idea is simple even though it sounds laborious. Instead of one profile for a whole episode, you cut the episode into stretches where the noise holds roughly still, and you give each stretch its own profile. Everything below uses plugins and actions that ship with REAPER.

1. Mark where the noise changes

Play through and drop a marker each time the background shifts. Listen in the gaps between sentences, not under the speech, because the gaps are where the change is audible. Most episodes have fewer zones than people expect. A fridge that cycles twice in an hour gives you five zones, not fifty.

Put the split point inside a pause wherever you can. A join that lands in the middle of a word is much harder to hide than one that lands in a breath.

2. Split the item at those points

Move the edit cursor to each marker and press S, which splits the selected items at the edit cursor. Work through the markers until the one long recording is a row of shorter items, one per noise zone.

3. Give each zone its own ReaFIR

There are two ways to do this and they are equally valid.

• Item FX. REAPER lets a single media item carry its own effects chain, separate from the track. Open the FX chain for that item’s take, add ReaFIR, and build the profile from a gap inside that item. Each item then keeps its own setting.
• One track per zone. If item FX feels fiddly, move the items for each noise condition onto their own track and put one ReaFIR on each track. Two or three tracks usually covers an episode, and everything stays visible.

Either way the procedure inside ReaFIR is the same for each instance. Set Mode to Subtract, start playback over a noise-only moment inside that zone, tick Automatically build noise profile, let it run for a second or two of noise, then untick it so the profile is frozen. If you need more reduction you can hold Ctrl and drag the profile upward, but the guide is explicit that raising it further produces more audible artifacts, so move it a little and listen rather than hauling it up.

4. Crossfade the joins

Two neighbouring items cleaned to slightly different degrees will not have identical noise beds, and an abrupt change in the noise bed is more noticeable than the noise itself. Overlap each pair of items by roughly 20–50 ms so REAPER crossfades between them, or draw a short fade on each side of the join. If overlapping does not produce a crossfade, auto-crossfade is probably switched off in Options.

5. Match the levels, then check the joins alone

Different amounts of subtraction change the perceived brightness and loudness of the voice, not just the noise. Play across each join with your eyes off the screen. If you can hear where it is, the usual culprits are too much reduction on one side or a level step between items, both of which are quicker to correct than to live with.

Once you are happy, consider rendering the processing into the items so the CPU cost goes away. REAPER can apply take effects to items and leave you the processed audio as a new take, with the original still there underneath.

When splitting is the wrong tool

Zones only exist if the noise holds still for a while. Traffic on a busy road, wind, rain and a room with people moving around in it do not give you zones, they give you continuous drift, and cutting that into forty items produces forty joins to hide for very little gain.

Two honest options for drift:

• Profile the quietest noise and accept the rest. Build the profile from the quietest stretch. You remove less, but you never over-subtract, so the voice stays intact and the leftover noise is even rather than lumpy. Even noise is far less distracting to a listener than noise that comes and goes.
• Use a denoiser that does not need a profile. Learning denoisers such as the open-source RNNoise estimate the noise from the signal as it goes rather than from one captured sample, which is a better mechanical fit for noise that will not sit still. They are also blunter, and they tend to make speech sound processed. There is a comparison of the free noise reduction plugins worth having for voice, including what each one actually does under the hood.

What this will not fix

• Noise that changes inside a sentence. A car passing while someone is mid-answer cannot be split out, because there is no gap to split at and no gap to profile from.
• One-off sounds. A door, a cough, a chair creak or a phone buzz is an event, not a noise floor, and subtraction will not touch it cleanly. Those are edited out by hand, which is covered in how to remove a cough or one-off noise in REAPER.
• Noise as loud as the voice. If the background is competing with the speech rather than sitting under it, no amount of subtraction separates them without wrecking the voice.
• Room echo. Reflections are not noise. They arrive with the voice and share its character, so a noise profile has nothing to subtract.
• A hiss that is coming from your gear. If the quiet gaps hiss identically on every recording you have ever made, the noise is probably your microphone or interface rather than the room. That is worth checking before you edit anything, and this is how to tell which one it is.

Stop it happening next time

This is the one genuinely cheap fix, and it is a scheduling decision rather than a technical one.

1. Turn the machine off for the recording. Air conditioning, fans, fridges and heat pumps can usually be switched off for an hour. A room that gets slightly uncomfortable beats an hour of editing.
2. Record a minute of room tone. Before or after the session, record the room with nobody talking. It gives you a clean profile source for every zone, and it gives you material to patch gaps with.
3. Get closer to the microphone. Halving the distance raises the voice relative to everything else in the room. This is the single most effective change most home recordists can make, and it costs nothing.
4. Record at a consistent time. If the noise is traffic, the same slot each week gives you a noise floor you can plan for instead of one you discover in the edit.

If it still sounds wrong

1. You reduced too much. Drop the profile back down. Leftover noise is almost always less objectionable than a processed voice.
2. The profile came from a gap with a breath in it. Breaths and mouth noise in the sample get subtracted from the speech as well. Find a cleaner second of noise and rebuild.
3. The zones are wrong. If a zone still sounds uneven, the noise changed inside it. Split again where it moves.
4. It is hum, not hiss. A fixed low pitch rather than a broad shhh is mains hum and has a narrower fix. See how to remove hum from a recording in REAPER.
5. Nothing is working. If the noise is loud, constant and broadband, re-recording a few minutes is often faster than another hour of editing. That is a legitimate answer, not a failure.