How to Fix a Quiet Podcast Recording Without Making It Sound Worse

The interview is great and the recording is whisper-quiet. Turning it up is easy; turning it up without also turning up the hiss, the room echo and every mouth click is the real job. This guide gives the repair order that works: fix balance first, then dynamics, then loudness — never the other way round.

When the repair is done, the final step is one click in the loudness normalizer, which lands the file on the platform target and proves it with a report.

Why Recordings Come Out Quiet

Quiet files usually come from cautious gain (better than clipping), a microphone too far from the speaker, or an interface set for a louder voice than the guest's. The recording itself is often fine — the detail is there, just low. What is also there, at the same low level, is the noise floor. Every decibel of gain raises both together, which sets the limit on how far simple gain can go.

A few causes recur so often they are worth naming:

  • The "safe gain" habit. Hosts are told — correctly — that clipping is unfixable, so they record 15 dB too low to be safe. They traded a rare disaster for a guaranteed problem.
  • The guest on a laptop mic. The built-in microphone sits a metre from the guest and hears the room as loudly as the voice.
  • A new voice, an old setting. The interface was set for the host's booming voice; the guest speaks softly, and nobody watched the meter.
  • Double-track drift. Each side recorded locally at sane levels, but one side exported with a fader down.

None of these ruin a recording. All of them share one property: the speech and the noise are both present, at a fixed distance from each other. Your repair budget is that distance — the signal-to-noise ratio — and every technique below is about spending it wisely.

First, Read the Numbers

Before touching anything, measure the file and write down three numbers: integrated LUFS, true peak, and — informally, from the waveform or a noise-floor reading — how far the hiss sits below the speech. Those numbers place you in the table further down this page and tell you whether you hold a routine repair or a rescue job. Guessing wastes the afternoon; measuring takes a minute.

Listen once, on headphones, at honest volume, and note the timestamps of the worst moments: the quietest passage that still matters, the loudest outburst, any hum, echo or handling noise. The repair order below is worked against that list, not against the whole file in the abstract.

The Repair Order

  1. Trim and clean first. Cut long silences and the worst noises before any gain, so meters read the real programme.
  2. Level the speakers. If one voice is 8 dB under the other, raise that voice's clips, not the whole file. Balanced voices are what loudness range measures, and balance is fixed by hand, not by normalization.
  3. Compress lightly. A gentle pass (2:1) pulls the quiet words up toward the loud ones, so the final gain has less work to do.
  4. Reduce noise only if you must. Noise reduction before gain; heavy-handed reduction after gain sounds watery. A little hiss at podcast loudness beats artifacts.
  5. Normalize to the target. Now measure and normalize to -16 LUFS for podcasts (the full workflow is in the -16 LUFS guide), with the true-peak ceiling from the true peak guide holding the loudest moments.

Why this order and no other? Each step assumes the ones before it. Gain applied before levelling raises the loud speaker as much as the quiet one. Compression before cleaning squeezes the noises into the programme where they are harder to remove. Noise reduction works best at the level the noise was recorded at, not after it has been amplified and compressed. And normalization is a measurement of the finished file — anything done after it invalidates the number. Do the steps in order and each one is easy; scramble them and each one fights the last.

A Worked Example: The -29 LUFS Interview

A 35-minute interview measures -29.3 LUFS, true peak -8.2 dBTP, and the hiss sits about 45 dB below the louder speaker — a healthy gap. The guest is roughly 6 dB under the host.

  • Trim: two long silences cut; the measurement now reflects the programme, -28.9 LUFS.
  • Level: the guest's clips come up 6 dB to match the host. The file now reads about -24 LUFS overall, and — more important — sounds like a conversation instead of a monologue with mumbling.
  • Compress: a gentle 2:1 pass lifts the quietest phrases toward the average. The needed gain drops further.
  • Noise: a light pass only, because at the final level a trace of hiss will sit under speech. No watery artifacts.
  • Normalize: two-pass to -16 LUFS, ceiling at -1 dBTP. Final report: -16.1 LUFS, -1.0 dBTP.

The hiss is audible if you solo the silences and listen for it. In the programme, under voices, at podcast loudness, listeners will never report it. That is the honest standard: not silence in the gaps, but a show nobody complains about.

How Far Can Gain Go?

Measured loudnessSituationPlan
-20 to -24 LUFSNormal quiet recordingSteps above; clean result expected
-25 to -32 LUFSVery cautious gainWorkable; listen for hiss after normalization
Below -32 LUFSNear the noise floorConsider re-recording the worst parts; gain will surface every flaw

The real variable behind the table is the gap between speech and noise, not the LUFS figure itself. A -30 LUFS recording made in a quiet room with a decent microphone can repair beautifully, because its noise sits 50 dB down. A -24 LUFS recording made next to a busy road may resist everything, because its noise is only 25 dB down and rises with every decibel of gain. Read the table as a guide, then trust the gap you measured.

When a file falls in the bottom row, triage honestly. If the content is a one-time interview, do the repair, accept the hiss, and publish — a flawed recording of an unrepeatable conversation beats a perfect recording of nothing. If it is your own solo material, re-record: twenty minutes now saves the repair and sounds better. If only parts are bad, re-record or patch those parts and level them in.

What Not to Do

Do not stack "make louder" effects, and do not normalize before editing — every cut changes the loudness you measured. Do not judge the result on laptop speakers; check on earbuds at normal volume, where your listeners live. And do not skip the final measurement: the report is the difference between "sounds louder to me" and "measures -16 LUFS," which is the number the platform sees — see what LUFS means.

Three more traps deserve names:

Do not confuse quiet with broken. A low file with a clean noise floor is a gain problem, fully fixable. Producers abandon good interviews because the waveform "looks small." Waveforms show peaks; they say nothing about how much clean signal is waiting to be raised.

Do not denoise aggressively to earn more gain. Heavy noise reduction buys a few decibels and pays with swirling, underwater artifacts that listeners find stranger than hiss. Hiss reads as "room"; artifacts read as "broken."

Do not fix it in the platform's hands. Uploading quiet and hoping the service turns you up hands the repair to a plain gain stage with no levelling, no ceiling care and no judgement — and some platforms, YouTube among them, will not turn you up at all.

Troubleshooting

The repaired file pumps — levels breathe up and down. The compressor is working too hard, usually because it was asked to supply most of the missing loudness. Back the ratio down, slow the attack, and let levelling and final gain do more of the work. Gentle stages in sequence beat one heroic stage.

The voice sounds boxy or underwater after noise reduction. That is over-reduction. Reduce less, or restrict the reduction to the gaps between phrases where noise is exposed and leave the speech itself alone. The goal is a quieter floor, not a silent one.

One channel is much quieter than the other. In a stereo file with one speaker per side, level the channels before anything else, or collapse to mono and treat it as one voice. Do not let the loudness step average away a channel problem — the report will look fine and the episode will still lean.

After all the steps, the report still shows -18 instead of -16. Something is holding peaks — usually a few loud moments forcing the ceiling. Find them (the report's true peak and your timestamp notes will agree), soften just those moments by a decibel or two, and re-run the final normalization. Targeted fixes beat global force.

Next Time: Recording So You Do Not Have to Repair

Repairs are for accidents; a small routine prevents most of them. Set gain so normal speech peaks around -12 to -6 dB on the recorder — high enough to bury the noise, low enough to survive a laugh. Get the microphone within a hand's width of the mouth; distance is the single biggest cause of quiet, roomy recordings. Record a ten-second test and listen on headphones before the real take, every time, including with regular guests — rooms and laptops change. And keep the raw recording: every repair in this guide starts from the quietest, cleanest original you have, and a file that was normalized, compressed and re-exported last year has already spent most of its repair budget before you begin.

Frequently Asked Questions

Can I just normalize a very quiet file and skip the rest? You can, and sometimes — clean room, one speaker, file around -24 LUFS — the result is perfectly good. The repair order earns its keep when the file is uneven or noisy. Measure first; let the numbers and one honest listen decide how much of the order you need, and skip with a clear conscience the steps your particular file does not call for.

Will turning up a quiet recording always make it hissy? Gain raises the noise floor by exactly the gain applied. Whether listeners hear it depends on the gap: noise 45 dB under speech stays masked under speech after +13 dB of gain; noise 25 dB under speech becomes a presence. Compression before gain genuinely helps here, because it raises the quiet speech toward the loud speech without touching the silences as much.

What if only the guest is quiet? That is a balance problem, not a loudness problem — step 2 of the repair order. Raise the guest's clips to match the host, then treat the combined file normally. Never normalize the whole file harder to rescue one voice; the host will end up limited and distorted while the guest is still soft.

The recording has echo — can loudness work fix it? No. Echo is baked into the voice itself; gain and compression only present it more clearly. Mild cases can be softened with dedicated de-reverb tools before the repair order begins. Severe cases are a re-record decision, kindly but firmly. This is the one common flaw with no loudness-side cure at all.

How quiet is too quiet to save? When the speech sits within about 20 dB of the noise, every cure audibly damages the patient. Below that line, publish only if the content is unrepeatable — and say nothing about it; most listeners are kinder than producers fear.

References

← All guides