How to Remove Background Noise From Audio

Steady background noise - hiss, a computer fan, air conditioning, distant traffic, mains buzz - is the most fixable problem in audio, because it is constant and predictable. Give a denoiser a short sample of the noise on its own and it can build a profile of exactly what to remove, then subtract it from the whole recording while leaving the voice largely intact. The trick is restraint: push it too far and you trade the noise for a watery, robotic artifact that is worse than what you started with. Aim for quieter and natural, not silent and processed.

Why steady noise is easy and sudden noise is not

Noise reduction works by prediction. A fan or a hiss floor sits at a stable level with a stable spectrum, so the tool can learn its fingerprint from a moment where nobody is talking and confidently pull that fingerprint out everywhere. That is why constant broadband noise responds so well.

Sudden, one-off sounds - a door slam, a cough, a chair creak - are the opposite. They do not repeat, so there is no profile to learn, and a broadband denoiser will barely touch them. Those need spectral editing or a gate, not noise reduction. Knowing which kind of noise you have decides which tool you reach for. If yours is a single-frequency electrical buzz rather than broadband hiss, treat it as hum and read how to remove hum from audio instead.

Step one: capture a noise profile

The single most important step is finding a clean sample of the noise by itself. Almost every recording has a second of "room tone" before the first word or after the last - just the noise, no voice. Select that region and let the denoiser learn it. That profile tells the tool the precise shape of what to remove, which is why profile-based reduction sounds so much cleaner than a blind, one-size-fits-all filter.

If there is genuinely no clean gap anywhere, use the quietest breath or pause you can find, or fall back to a modern learning denoiser that estimates the noise floor automatically. Sauce's Audio Repair handles this for you - it profiles the noise and applies reduction in one pass, so you can hear how clean a take gets before deciding whether to do anything more.

Step two: denoise gently

Set the reduction amount, then treat it like turning up a dimmer, not flipping a switch. Bring it up until the noise drops away, then keep going one notch too far on purpose so you can hear the failure mode, then back off until it disappears. That failure mode has a signature: a swirly, underwater, "chirpy" quality, sometimes called musical noise, and a thin, robotic edge on the voice. The moment you hear it, you have gone too far.

It is almost always better to leave a faint, natural noise floor than to scrub the recording completely and be left with artifacts. The ear forgives a little consistent hiss far more readily than it forgives warbling. If 8 dB of reduction sounds clean and 16 dB sounds processed, take the 8.

Step three: gate the gaps (optional)

Once the broadband noise is reduced, a gentle gate or expander can clean up the silences between phrases, where any remaining noise is most exposed. Set a slow release so it closes smoothly rather than chattering on and off. This is a finishing touch, not the main event - do the denoise first, because gating alone does nothing about the noise sitting underneath the words themselves.

What can and cannot be fixed

• Steady hiss, fan, air conditioning, mains hum - removes cleanly and convincingly. This is the ideal case.
• Constant traffic or crowd rumble - reduces well, though heavy low-frequency energy may need a high-pass filter as well.
• Intermittent noise - dogs, doors, keyboard, coughs - denoise barely helps. Use a spectral editor to paint out individual events, or re-record the affected lines.
• Noise louder than the voice - you can improve intelligibility, but you cannot fully separate the two. When the signal-to-noise ratio is that bad, some noise always rides along with the words.
• Wind rumble and handling thumps - a high-pass filter around 80 to 120 Hz does more than any denoiser, because that energy lives below the voice.

A quick step-by-step

• Find a second of noise with no voice and capture it as the noise profile.
• Apply denoise and raise the amount only until the noise recedes, then back off the instant you hear swirling or a robotic edge.
• High-pass any low rumble around 80 to 120 Hz that sits below the voice.
• Add a gentle gate to clean the gaps between phrases if the silences still hiss.
• Compare against the original at matched volume - if the "clean" version sounds thinner or weirder, dial the reduction back.

If the noise is one problem among several, treat it as a stage in the wider cleanup covered in how to clean up a vocal recording, and always fix noise before you shape tone.

Prevention beats repair

The cleanest denoise is the one you never have to do. Get close to the mic so the voice dominates the noise floor. Turn off fans, air conditioning and anything that hums before you hit record. Record in the quietest, softest room you have - a closet full of clothes beats a bright empty room. And always capture a few seconds of room tone at the top of every session, because that clean noise sample is the thing that makes repair painless later.