Reduce Background Noise in a Recording Without Damaging Speech

Capture first: the best noise reduction is the noise you never recorded

Every noise reduction algorithm is a repair job, and every repair job costs something, even a careful one. Before you reach for a cleanup tool, it is worth spending thirty seconds on the source: close a window, turn off a fan or air conditioner, move a laptop's cooling fan away from the microphone, or simply sit closer to the mic so your voice sits further above the room noise. A recording that starts cleaner needs less aggressive processing later, and less aggressive processing means fewer artifacts in the final result.

This matters more than it might seem, because there is a real limit to how much a cleanup step can accomplish once noise is baked into a recording. Reduction settings can pull the noise floor down considerably, but they cannot separate a voice and a competing sound perfectly when the two overlap in both time and frequency — the cleanest fix is always to keep the noise out of the recording in the first place.

How the noise profile actually works

The noise remover works by analyzing your audio in short overlapping windows (a 2048-point spectral analysis) and comparing each window against a profile of what the unwanted noise looks like across the frequency spectrum. Where a frequency matches the noise profile, it gets reduced; where it does not, it is left alone. That is why the quality of the profile matters more than almost any other setting.

Auto mode builds this profile from the quietest sections of your file automatically, which works reasonably well for steady, consistent noise. Learn from a selection lets you mark an exact stretch of pure noise — a couple of seconds of room tone before anyone starts talking, for example — and tends to produce a more accurate profile than Auto when the noise is distinctive, like a specific hum or a fan with an uneven pitch.

Why conservative reduction beats maximum reduction

It is tempting to push reduction to 100% and be done with it, but spectral noise reduction gets audibly worse, not just more effective, as you push it harder. Over-subtracted audio develops a characteristic "watery" or metallic quality, sometimes called musical noise, as isolated frequency bins pop in and out. Smoothing helps soften that artifact, but the more reliable fix is simply not asking for more reduction than the recording needs.

The Listen to removed noise option is the fastest way to check your settings honestly. It plays back exactly what the tool subtracted. If that removed signal is just hiss, hum, or fan noise, your settings are working. If you can make out words or vocal tone in it, the reduction is too aggressive and is eating into the speech itself — dial it back.

Finishing the cleanup: EQ and loudness

Noise reduction handles the unwanted signal, but it will not make a thin or uneven recording sound polished on its own. Once the background noise is under control, a light touch with the EQ — the Podcast voice preset is a reasonable starting point — can add presence and clarity back to a voice that noise reduction has slightly dulled. From there, set a consistent loudness target with the volume booster so the finished file plays back at a predictable level; see the guide on normalizing podcast loudness for target numbers and how to check true peak.

What noise reduction cannot fix

Spectral noise reduction is built for steady, predictable noise — hiss, hum, a fan, an air conditioner, room tone. It is honestly not a good tool for a second person talking in the background, overlapping conversation, or any other noise that shares the same frequency range and pattern as speech itself; there is no profile that separates one voice from another this way. It also will not undo clipping or distortion from a recording that was too loud to begin with, and it will not fix a room that sounds boomy or reverberant, beyond what the rumble high-pass filter can trim off the low end. Recognizing which category your problem falls into saves time before you start turning up the reduction slider looking for a fix that is not there.

If a recording has more than one of these problems at once — say, fan noise plus a boomy room — treat them separately rather than expecting one pass to fix everything. Clean the steady noise first, check the residual, and only then decide whether the remaining character of the recording needs a different kind of fix, such as EQ, rather than more reduction.