How to Remove Background Noise From Audio or Video (2026)

The three kinds of background noise

“Remove the background noise” sounds like one problem, but there are really three, and they don’t all respond to the same fix:

• Broadband noise — steady hiss or air. Cheap mics, high gain, fans, air conditioning, distant traffic that blurs into a wash. This is the most common and, fortunately, the most fixable.
• Hum — a specific low-frequency tone, usually 50 or 60Hz from mains electrical interference, ground loops, or nearby equipment. Narrow and tonal.
• Intermittent noise — keyboard clicks, a dog barking, a door, a cough. Sporadic and unpredictable, which makes it the hardest to remove cleanly.

Knowing which one you have tells you what will and won’t work.

Why EQ and noise gates aren’t enough

The two classic manual tools both have a fatal flaw on voice:

A noise gate silences audio below a volume threshold. It works when there’s a clear gap between “loud voice” and “quiet noise” — but the moment someone speaks, the gate opens and all the noise comes back with the voice, because the noise was there the whole time, just now under the speech. Gates clean up the pauses and do nothing for the words.

EQ (cutting frequencies) can kill a narrow hum beautifully, because hum lives at one frequency. But broadband hiss overlaps the same frequencies as the voice itself — cut enough to remove the hiss and you make the voice sound thin and muffled. You’re throwing away the baby with the bathwater.

This is exactly why AI denoising exists: it’s trained to tell voice apart from noise even when they occupy the same frequencies at the same time, which no gate or EQ curve can do.

How AI denoising actually works

Instead of a fixed rule (“cut everything below X volume” or “cut this frequency”), an AI denoiser has learned what human speech looks like versus what noise looks like, from huge amounts of examples. It processes the audio and reconstructs the clean voice while suppressing everything it recognizes as noise — including intermittent sounds a gate would miss and broadband hiss an EQ would smear.

To use it, run your file through the audio denoiser. One useful detail: it accepts video files too, not just audio — it cleans the audio track in place and hands back the video, so you don’t have to demux and remux yourself. Upload, process, download.

Realistic expectations

AI denoising is genuinely good now, but it’s not magic, and honesty here saves you frustration:

• Steady hiss and hum → nearly perfect removal in most cases.
• Moderate room noise → strong improvement, occasionally a faint “underwater” artifact if the noise was very loud relative to the voice.
• Intermittent clicks and bumps → mostly removed; very loud transients might leave a trace.
• Music or speech behind your speech → this is the hard limit. A denoiser suppresses noise, but a music bed or a second talker is structured audio, not noise. It can’t cleanly pull your voice out from over a song.

That last case needs a different tool entirely. If what’s “in the background” is music, you don’t want a denoiser — you want source separation. The vocal remover and stem splitter are built to split a mix into separate voice and instrument tracks, which is a fundamentally different operation from noise reduction.

And if the “noise” you actually want gone is the speaker’s own ums, uhs, and dead air, that’s remove filler words — a timing edit, not a spectral one.

Recording cleaner audio in the first place

Every minute spent on the recording saves ten in cleanup. You don’t need a studio:

Get closer to the mic. This is the single biggest lever. Doubling your distance from the mic roughly quarters the voice level relative to the room, so the closer you are, the more your voice dominates the noise. Six inches beats two feet dramatically.

Treat the room on a budget. Hard walls bounce sound and add echo (which is a separate problem from noise — see the echo remover for that). Soft things absorb it: record in a room with a carpet, curtains, a couch, or even hang a blanket. A closet full of clothes is a famously good free vocal booth.

Kill the obvious sources. Turn off the fan, the AC, and anything with a hum before you hit record. Thirty seconds of prevention beats an AI reconstruction.

Record a few seconds of silence. Some workflows use a “noise profile” — a clean sample of just the room tone — to key the removal. Even if your tool doesn’t need it, having a silent reference helps you hear what the noise actually is.

The practical sequence

Put it together and the workflow is:

1. Record as clean as you reasonably can (close mic, soft room, sources off).
2. Run the file through the audio denoiser — handles video too.
3. If there’s music behind the voice, use the vocal remover instead, because that’s separation, not denoising.
4. If the problem is echo, not noise, that’s a different pass.
5. If it’s the speaker’s filler words, tighten with remove filler words.

Match the tool to the actual problem and the results are far better than reaching for one “clean up my audio” button and hoping. Noise, music, echo, and filler are four different jobs — and each has a tool that’s actually built for it.