How to Remove Crowd Noise from Audio and Video (Cafe, Market, Event)
Lower distant crowd noise behind a voice, and know which nearby voices will stay. Real before/after samples with scores, a free 30-second preview, and when to reshoot.
You can lower distant crowd noise behind a voice, but a nearby person whose words you can make out is speech too, and no noise remover can fully separate the two. The wash of a market or a busy room drops well. The guy at the next table finishing his sentence usually doesn't.
The fastest way to find out which kind you have is to upload the clip to the EraseNoise video noise remover and play the free 30-second preview against the original. If the crowd turns into a soft bed under your voice, keep it. If you can still follow someone else's sentence, read the section on what you can't fix before you spend anything.
Key Takeaways
- Crowd noise is many voices at once, so it sits in the same frequencies as the person you want to keep. That makes it harder to remove than a fan or traffic.
- Distant babble lowers well. In our real laughing sample, the background score rose from 1.67 to 4.03 out of 5.
- Near, intelligible voices stay. In the same samples the speaker's naturalness stayed under 3 out of 5, so judge the voice, not the silence.
- Audacity's manual says its Noise Reduction isn't suitable for irregular noise such as an audience, so a noise profile won't fix a crowd.
What you'll hear in a crowd recording
Listen once before you touch anything, and sort the noise into these groups. Each one gets a different fix.
| Sound | Steady or sudden | What a cleanup can do |
|---|---|---|
| Distant babble: a market, a hall, a busy room | Steady, many voices blurred together | Drops a lot. Usually the easiest part |
| A nearby person you can understand | Sudden, one clear voice | Often stays. It is speech, like yours |
| Laughter, shouts, cheers | Sudden and loud | Gets quieter, but the voice under it can sound strained |
| Applause | Sudden bursts between sentences | Use the applause remover, or cut the bursts |
| Music from speakers or a stage | Steady with a beat | A different job. See the cafe vlog guide |
The first row is what most people mean by crowd noise, and it's the part you can expect to fix. The second row is why "remove people talking in the background" doesn't have a one-click answer.
Photo: Jorge Láscar, CC BY 2.0, via Wikimedia Commons.
Why crowd noise is harder to remove than a fan
Crowd noise is hard because it's made of the same material as the voice you're keeping. A fan is one steady sound with its own spectrum, so a tool can learn it and subtract it. A crowd is dozens of voices whose energy sits right where speech sits, and it changes every second.
That is why classic noise profiles fail here. The Audacity manual says Noise Reduction works on constant background sounds such as hum, hiss, or fan noise, and isn't suitable for irregular background noise such as traffic or an audience. It also warns that a satisfactory removal may be impossible when the noise sits in the same frequencies as the voice (Audacity Manual: Noise Reduction, retrieved 2026-10-26). A crowd is both.
Speech models do better on distant babble, because they're trained to keep one foreground voice and drop everything else. They can still only guess when a second voice is close and clear. To the model, that voice looks like the one you want.
Fixing a crowd clip you already shot
Start with the free fixes in your editor, then use a speech model for the steady wash. The two work best together.
Free fixes in any editor
- Cut or duck the loudest seconds. A cheer, a shout, or a burst of laughter between your sentences is an edit, not a cleanup job. Split the clip around it and lower that piece or remove it.
- Fill gaps with room tone. If a cut leaves dead silence, paste a second of the crowd from another pause. A soft, even crowd sounds more natural than a crowd that switches on and off.
- Lay music under what's left. On a vlog, a quiet music bed hides leftover babble better than pushing the cleanup harder. Use music you have the rights to.
Don't reach for a noise print here. A print of a crowd describes one moment of a sound that never repeats.
Using EraseNoise on the steady wash
- Upload the video to the video noise remover, or the audio file to the audio noise remover. Video can be MP4, MOV, M4V, or WebM up to 2 GB. Audio can be MP3, WAV, M4A, AAC, OGG, Opus, or FLAC.
- Listen to the free 30-second preview. It picks the stretch with the most sustained sound in the first 10 minutes, which in a crowd clip is usually the busiest one. No account is needed for 3 previews a day.
- Lower the strength slider if the voice turns thin. A crowd that drops to a soft bed often sounds better than a crowd that's gone, especially at a market or an event where the place is part of the story.
- Download only if your voice is clearer. Video comes back as MP4 with the picture copied as it is. Audio comes back as WAV. Full files use paid minutes, and uploads are deleted after 24 hours.
Hear it on real recordings
These are real recordings from the Microsoft DNS Challenge test set, processed by EraseNoise and scored with DNSMOS (overall and background quality) and NISQA (voice naturalness), each on a 1–5 scale. We haven't published a market or event recording of our own yet. These two are the closest: people laughing and yelling around a speaker.
People laughing behind a speaker. Background 1.67 → 4.03, overall 1.72 → 2.86, naturalness 1.22 → 2.94.
Original
Cleaned by EraseNoise
Yelling behind a speaker. Background 3.29 → 4.05, overall 2.16 → 2.79, naturalness 1.37 → 2.44.
Original
Cleaned by EraseNoise
Read the scores honestly. The background got much cleaner in both, but the voice stayed under 3 out of 5 for naturalness. For comparison, our street recording went from 1.04 to 4.32 on the same naturalness scale. Steady outdoor noise is a much easier job than other people's voices. All samples are on the noise removal examples page.
Editor fixes vs EraseNoise for crowd noise
| Job | Better first workflow |
|---|---|
| One cheer or shout between your sentences | Any editor: cut or duck it |
| A steady crowd wash under the whole clip | EraseNoise: upload, preview the busiest stretch |
| A crowd you want to keep a little of | EraseNoise with the strength slider turned down |
| Applause between speeches | The applause remover, or cut the bursts |
| Someone nearby talking over your words | Reshoot or re-record the line. See below |
Common mistakes
- Judging the pauses. A silent gap tells you nothing. Listen to your own sentences, because that's where the damage shows.
- Pushing for total silence. The voice goes first. Stop when your words start to sound watery or clipped at the ends.
- Using a noise profile. It describes one moment of a sound that never repeats.
What you can't fix
Some crowd recordings can't be cleaned into something natural, and it's cheaper to know that before you pay for a full file.
- A nearby voice you can understand. If you can follow the stranger's sentence, so can the model, and it may keep it.
- Someone talking at the same moment as you. Two voices on top of each other can't be pulled apart cleanly. Expect your own voice to suffer.
- A crowd louder than you. If the crowd drowns your words in the original, there isn't enough of your voice left to keep.
- Clipping. If a roar pushed the recording into distortion, those words are damaged. Noise removal can't rebuild them.
In those cases, re-record the line as a voiceover, or cut to a quieter take.
Next time: free ways to prevent crowd noise
- Get the mic close to your mouth. A clip-on or wireless lavalier a hand's width below your chin beats any camera mic. The closer you are, the smaller the crowd is by comparison.
- Put your back to the crowd. Most mics hear best in front of them. Face a wall or a corner and let your body block the noise.
- Pause during the loud moments. Wait for the cheer, the announcement, or the laugh to finish. A three-second wait saves an edit.
- Record 10 seconds of the room. Room tone gives you something to paste over cuts so the crowd doesn't switch on and off.
- Step aside for the talking parts. Film the crowd for the visuals, then say your lines from the edge of it.
Next step
If the crowd problem is really applause between speeches, use the applause and clapping remover. If the crowd was in a cafe, the music on the speakers is a separate problem that can also get a video claimed on YouTube. The cafe vlog guide covers both. For street, wind, and car scenes, start from the vlogger's guide to clean audio.
When the next crowd clip is ready, upload it to the video noise remover and listen to the busiest 30 seconds before you decide.
Frequently asked questions
Can you remove people talking in the background?
Distant chatter, yes, mostly. It blends into a wash that a speech model can lower. A nearby person whose words you can make out is speech too, and it often stays. If someone is talking over your sentence, re-record that line.
How do I remove crowd noise from a video?
Upload the video to a speech-focused noise remover and compare a preview of the busiest stretch with the original. With EraseNoise the 30-second preview is free and the picture isn't re-encoded. Cut or duck any shouts and cheers between your sentences in your editor.
Why does crowd noise removal make the voice sound strange?
The crowd and the voice share the same frequencies, so the tool has to guess which parts belong to you. Pushed too far, it takes some of your voice along with the crowd. Turn the strength down and leave a little crowd in.
Can I remove applause from a recording?
Yes. Applause usually comes in bursts between sentences, so you can cut it or lower it in an editor. For applause under speech, use a tool made for it, such as the EraseNoise applause remover.
Should I remove all the crowd noise from an event video?
Usually not. A little crowd tells viewers where you are, and removing all of it is where voices start to sound processed. Aim for a crowd that sits clearly under your voice.
Sources
- Audacity Team, Audacity Manual: Noise Reduction, retrieved 2026-10-26.
- Microsoft, DNS Challenge test set, source of the laughing, yelling, and street recordings, retrieved 2026-10-26.
- EraseNoise, noise removal examples: DNSMOS and NISQA scores for each sample.
