Can background noise be removed completely?

Steady noise such as hiss, hum and fans can be reduced a lot. Clicks, crowds and other voices cannot be removed without touching the speaker's voice, so expect to reduce them, not erase them.

Why does my voice sound watery after noise reduction?

The reduction is set too high. Strong noise reduction removes parts of the voice that sit near the noise. Lower the amount, apply it to the voice track only, and compare against the original while you listen.

Should I remove noise before or after normalising loudness?

Before. Normalising changes the overall level, and raising a noisy track raises its noise too. Reduce noise, balance the mix, then normalise last.

How do I stop keyboard sounds in my videos?

Fix it at the source: mount the mic on an arm away from the desk, use a dynamic mic close to your mouth, or use a quieter keyboard. Noise reduction works poorly on short sharp clicks.

Guide

How do I remove background noise from a video?

Updated 18 October 2026

In short: work out what kind of noise it is, because steady noise (hiss, hum, fans) can be reduced a lot and sudden or voice-like noise (keyboard clicks, a crowd, other people) cannot. Apply noise reduction to the voice only, use the least that works, because too much makes the voice sound watery, and normalise loudness last. The cheapest fix is a cleaner recording: get the mic closer and lower the gain.

What kind of noise is it?

Noise	What it sounds like	Usual cause	What works
Hiss	A steady "shhh" under everything.	Gain turned up too far, a cheap mic or interface.	Lower the gain and move the mic closer. Then noise reduction.
Hum or buzz	A low steady tone, sometimes with higher overtones.	Mains electricity getting into the cables: ground loops, power cables lying across mic cables.	Fix the cabling. A notch filter set to the mains frequency (50 or 60 Hz, depending on your country) removes what is left.
Fans, air conditioning, a PC	A low, steady rumble or wash.	Machines in the same room as the mic.	Move the mic or the machine. A high-pass filter plus noise reduction handles the rest.
Keyboard and mouse clicks	Short sharp taps while you talk.	A mic on the desk, a loud mechanical keyboard.	A mic arm, a dynamic mic, softer switches. Noise reduction barely helps because the sound is not steady.
Crowd, other voices, a TV	Speech-like sound behind your voice.	A room with other people, a street, an event.	Record closer to the speaker. AI voice isolation can help, but expect artefacts.

The rule behind the table: noise that stays the same from one second to the next can be learned and subtracted. Noise that changes, or that sits in the same range as speech, cannot be separated without also touching the voice.

Record cleaner first

Mic distance. The closer the mic is to your mouth, the louder your voice is compared with everything else in the room. That one change helps more than any filter.
Gain. Set it so your loudest moments stay clear of clipping and the meter moves comfortably. If you record too quietly and boost later, you raise the hiss with it.
Mic type. Dynamic mics pick up less of the room than condenser mics, which helps in an untreated room.
Separate tracks. If you can, record your mic on its own track from the game and chat. Then you can clean the voice without touching the rest.
Room tone. Record a few seconds of the room with nobody speaking. Some tools use it to learn the noise.

What noise reduction does, and what it breaks

Most noise reduction works on the frequency content of the sound. It estimates the noise and turns it down in the places where only noise is present. FFmpeg's documentation for its FFT denoiser, for instance, has a setting for how many decibels of reduction to apply and one for the level of the noise floor (FFmpeg filters documentation, read 18 October 2026). Newer tools use trained models that separate a voice from the rest of the sound.

Both can leave marks on the voice:

A watery or bubbling quality when the reduction is too strong.
Dull consonants. The soft edges of "s", "t" and "f" sit near the noise and get turned down with it.
A dead room. Taking all the ambience out makes the voice sound pasted on.
Pumping. The noise comes and goes with your speech because the reduction is switching on and off.

Use less than you think you need, and switch the effect on and off while you listen. If you can only hear the difference when you listen for it, it is enough. Check the result on a phone speaker, not only on headphones.

Loudness normalisation

Peak level and loudness are different things. A peak is the loudest instant. Loudness is how loud the whole clip feels. Normalising loudness brings clips to a consistent level so viewers do not have to change the volume between one clip and the next. FFmpeg's loudness filter, for example, implements the EBU R128 method and works from three targets: integrated loudness, loudness range and maximum true peak (FFmpeg filters documentation, read 18 October 2026).

There is no target number here for TikTok, Instagram or YouTube. We could not open an official help page that states one for short-form video, and we would rather not repeat a figure from a blog. Check the platform's current guidance if you need an exact figure.

One caution for streams: normalising makes the loudest thing the reference. If the game is blasting and your voice is quiet, normalising keeps that balance. Fix the mix first, so the voice is clearly on top, and then normalise.

The order that works

Record clean. Close mic, sensible gain, voice on its own track if possible.
Cut first. Remove the parts you will not use, so you do not process them or let them influence the settings.
Remove rumble and hum. A high-pass filter for low rumble, a notch filter for mains hum.
Reduce noise on the voice only. Do it before music or effects are added, and keep it light.
Balance the mix. Voice on top, music and game audio below it. Music that ducks under the voice does this automatically.
Normalise loudness last. Anything done after it changes the level you just set.
Listen on a phone. Check the quietest word and the loudest laugh.

How to do it in RipClip

In the RipClip editor you can switch on Reduce noise and Normalize loudness for a clip. Voice enhance and noise removal are Premium features. Premium features can be used by anyone while editing, but exporting them needs a paid plan. Music ducks under voices, so the voice stays on top, and you can record a voiceover with 103 voice effects. RipClip clips Kick and Twitch VODs, highlights and clips, cloud-drive share links, direct video links and uploads. It does not take YouTube links; upload the file instead.

This is a quick clean-up for clips, not audio restoration. If a recording has loud crowd noise or heavy distortion, a dedicated audio repair tool will do better, and recording cleaner next time will do better still. RipClip does not record your game or screen.

Related: removing silences and pauses, adding captions automatically, where clean audio also gives more accurate captions.

Sources

FFmpeg filters documentation, section afftdn: an FFT-based denoiser with a noise-reduction amount in decibels and a noise-floor setting. Read 18 October 2026.
FFmpeg filters documentation, section loudnorm: EBU R128 loudness normalisation with targets for integrated loudness, loudness range and maximum true peak. Read 18 October 2026.
The advice on microphones, gain, noise types and the order of steps is common audio practice, not taken from a platform page or a study.