How to remove background noise from podcast audio without hollowing out the voice

Audacity's manual contains the sentence most noise-removal tutorials leave out: “Reducing noise usually results in some distortion. This is normal and there is nothing you can do about it.” To remove background noise from podcast audio, record a few seconds of the room with nobody talking, feed that to your editor as a noise profile, and apply the smallest reduction that makes the noise tolerable. That is the whole method. The reason it works better for some shows than others has nothing to do with which tool you pick and everything to do with what kind of noise you recorded.

Any podcast can be listed on PodAnswer for free, and a listed show gets a page of its own that can turn up when somebody searches the show's name, so take two minutes and claim the listing at https://podanswer.com/signup while your episode renders.

Noise you prevent at the microphone costs nothing to repair. Noise you remove in software costs you part of the voice.

What the software is actually doing to your voice

Audacity's Noise Reduction effect works in two passes. First you select a stretch of audio containing nothing but the noise and click Get Noise Profile, which measures the noise across frequency bands roughly 21.5 Hz wide at a 44,100 Hz sample rate. Then it subtracts that fingerprint from the whole track. Your voice occupies many of the same frequency bands as a furnace or a laptop fan, so every decibel you take out of the noise takes something out of the speech sitting on top of it.

The Audacity manual, updated 2 September 2026, puts a number on the restraint this calls for. The Noise Reduction control defaults to 6 dB, and the manual's instruction is to “use the lowest value that reduces the noise to an acceptable level,” with the warning that “higher values than necessary may make the noise even quieter, but will result in damage to the audio that remains.” The Sensitivity control also starts at 6.00 on a 0 to 24 scale, and the guidance there is the same: find the lowest value that works “without the introduction of artifacts.”

Those artifacts have a name and a sound. The manual describes “random bursts of very short tones at random frequencies, sometimes called ‘musical noise’,” and is blunt about the trade: they “can be even more objectionable than the original noise.” If a cleaned-up episode sounds like the host is speaking from inside a wind chime, that is a 24 dB reduction setting, not a bad microphone.

The three noises, and the one that software can fix

The reason noise removal fails for so many podcasters is that they point it at the wrong problem. Audacity's manual draws the line explicitly. The effect “can reduce constant background sounds such as hum, whistle, whine, buzz, and ‘hiss’,” and it “is not suitable for individual clicks and pops, or irregular background noise such as from traffic or an audience.”

Sort what you recorded into three piles.

Steady noise. Refrigerator hum, a computer fan, air conditioning, mains hum at 50 or 60 Hz, tape hiss, the faint whine of a cheap interface. This is what noise reduction was built for, and one gentle pass usually clears it.

Intermittent noise. A dog, a siren, a door, a chair creak, a neighbour's leaf blower that runs for ninety seconds in the middle of your best answer. No noise profile describes it, because it is not there most of the time. The fix is editorial: cut it, talk over it, or rerecord the line. Treating it with reduction damages the whole track to chase a sound that appears in four places.

Reflections. A bathroom-sounding room is not noise at all. It is your own voice arriving at the microphone twice, once directly and once off the wall behind you. Subtracting a profile of the room does nothing, because the thing you want gone is correlated with the thing you want to keep. Reverb needs a different tool, or a different room.

The manual is also honest about the ceiling: it says it “may be impossible to get a satisfactory removal when the noise is very loud” or when the noise is variable. A recording made next to a running dishwasher is not a software problem.

Twenty seconds that decide how well this goes

Audacity needs a minimum of 2,048 samples to build a profile, which at 44,100 Hz is roughly five hundredths of a second. That is the floor, not the target. A profile taken from a half-second gap between words includes the tail of your consonants and teaches the effect to subtract parts of your speech.

So record room tone on purpose. Before anybody talks, sit still for twenty seconds with the microphone running at the same gain and the same distance you will use for the episode. Do it at the top of every session, because the noise floor changes when the heating comes on, when a different laptop is in the room, or when you move the mic two inches. Studio engineers treat this as part of setup rather than as a rescue step, which is the same instinct behind getting a drum sound right in the room instead of in the mix, a subject the hosts of Recording Studio Rockstars work through at length in our answer page on https://podanswer.com/answers/how-do-you-get-a-deep-punchy-snare-sound.

Twenty seconds of clean room tone, saved as its own file, is also the fastest way to find out whether your problem is fixable before you spend an hour on it.

Hosted processing, and the number on the invoice

If you would rather not touch an effect panel, Auphonic runs noise and reverb reduction as a hosted service and publishes its free allowance plainly: “Auphonic is free for 2 hours of processed audio per month.” The free tier includes noise and reverb reduction and the filtering and enhancement stage, and it caps multitrack productions at twenty minutes. Two details are easy to miss on that pricing page. Free productions “come with a jingle” appended to the output, and billing runs on the duration of audio you send, with a floor: Auphonic states it will “charge at least 3min of credits per production to avoid spam attacks.” A show publishing two 45-minute episodes a month is already past the free allowance.

The arithmetic worth doing before you buy anything: two hours a month covers a weekly show only if episodes run under thirty minutes, and it covers nothing at all if you process the same file three times while experimenting. Process once, after you have made the editorial cuts.

The order that wastes the least audio

1. Fix the room. Close the window, switch off the HVAC for the recording, move away from the wall, put something soft behind the microphone, and get your mouth within a hand's width of a dynamic mic. Every one of these is free and permanent.
2. Record twenty seconds of room tone at the session's real gain and distance.
3. Edit first. Cut the intermittent noises, the restarts and the long pauses before any processing runs, so you are not spending reduction on audio that will not ship.
4. Build the profile from the room tone, not from a gap in speech.
5. Apply one pass at the default 6 dB. Listen on headphones to a sibilant-heavy sentence, which is where musical noise shows up first. Increase only if you must, and stop the moment the voice starts to sound thin or watery.
6. Do loudness last. Normalization after reduction, never before, because a loudness pass applied to a noisy file lifts the noise with the voice. The targets Apple and Spotify publish, and what they mean for your export, are in https://podanswer.com/blog/how-loud-should-a-podcast-be.
7. Keep the unprocessed master. If the next tool is better in a year, you will want the original file rather than the version you already subtracted from.

Two upstream decisions make all of this easier. Recording at a sensible sample rate and delivering at the right one is covered in https://podanswer.com/blog/what-sample-rate-for-podcast, and remote guests, who contribute most of the bad audio on most shows, get their own set of precautions in https://podanswer.com/blog/record-podcast-remote-guest.

Questions that come up

Can you remove background noise from podcast audio for free? Yes. Audacity's Noise Reduction effect is free and documented, and Auphonic processes two hours of audio a month at no cost, with a jingle attached to free output. Neither one can repair a recording where the noise is as loud as the voice.

How much noise reduction should I apply? Start at Audacity's default of 6 dB. The manual's own instruction is to use the lowest value that brings the noise to an acceptable level, because higher values damage the speech that remains.

Why does my podcast sound tinny or watery after noise removal? That is musical noise, the artifact Audacity's manual describes as random bursts of short tones at random frequencies. It means the reduction or sensitivity setting is too high. Undo, halve the setting, and apply once rather than twice.

Will noise reduction remove echo from a room? No. Echo is your own voice reflecting off hard surfaces, so it is correlated with the speech and a noise profile cannot separate it. Auphonic's reverb reduction is built for it, and soft furnishings work better than either.

Should I run noise reduction before or after loudness normalization? Before. Normalizing first raises the noise floor along with the voice and leaves the reduction stage a harder job.

Does a better microphone fix background noise? A dynamic microphone used close to the mouth rejects far more of the room than a condenser used at arm's length, so the upgrade that helps is usually about pattern and distance rather than price.

What a clean episode still does not do

Audio quality keeps listeners you already have. It does not bring you new ones, because nobody searches Google for a quiet furnace.

That part is a separate job, and it is the one PodAnswer does. We read your transcripts, find the questions people are already typing that your episodes answer, and publish each answer as its own page with the relevant excerpt quoted and timestamped and a link back to the episode. Growth is $149 a month for four articles, Network is $349 a month for twelve. What the service covers and what we need from you is at https://podanswer.com/for-podcasters, and both plans are at https://podanswer.com/pricing.

Spend the twenty seconds on room tone. Spend the rest of the hour on something a stranger can find.