In this guide
Decide what the finished recording must retain. To focus on a speaker, test the voice as the target and review Isolated sound. To preserve music and ambience, test the crowd as the target and review Background audio. Compare complete overlap and transitions; neither approach guarantees a clean separation.
To reduce crowd noise in an audio or video recording, first decide whether you need only a clear speaker or the wider sound of the event. For a speech, try isolating the voice. For a scene where music, footsteps or room ambience matter, try isolating the unwanted crowd and review what remains. If a single cheer occurs in a pause, a local volume edit may be enough.
Keep the original and test a short, difficult passage before processing the whole recording. Crowd chatter includes voices, claps and changing sounds, so a result that quiets the audience can also change the person or music you wanted to keep.
Choose a method for your recording
| Situation | First test | Listen for |
|---|---|---|
| Interview or speech over chatter | Target the spoken voice; review Isolated sound. | Quiet words and consonants when other people talk. |
| Event scene with music and applause | Target applause or crowd noise; review Background audio. | Music attacks, speech and ambience that may also change. |
| One cheer between sentences | Try a gradual volume dip in the pause. | Room tone and natural transitions. |
| Live performance with crowd under music | Look for the soundboard recording or individual tracks; compare isolation if unavailable. | Instrument and vocal detail throughout the arrangement. |
If there is a separate camera, lavalier or soundboard recording, compare it first. A closer or separately recorded source can preserve detail that is difficult to recover from a crowded mix. A soundboard feed may lack room atmosphere, so choose the source that serves the edit.
Test a speaker-focused result
- Choose a representative excerpt. Include a quiet phrase, the loudest crowd overlap and a little audio on both sides. Note the excerpt's original start time if it will go back into a video timeline.
- Upload the original. Open background noise removal. The interface supports MP3, WAV, MP4 and MOV files up to 50 MiB (52.43 MB). If the source is larger, export a short test from your editor.
- Describe the sound to keep. Try “spoken voice” or “the person speaking” and select Isolate sound. Review the resulting Isolated sound as the voice candidate. Descriptions are starting points, not guaranteed settings.
- Compare complete sentences. Listen at a similar playback level to the original. Check word beginnings, endings, interruptions and sections where several people speak together. The result may retain other voices or change the wanted speaker.
This approach can also reduce sounds you wanted in the scene. If retaining event music or applause matters, compare the crowd-targeted approach below instead of assuming the voice output preserves the whole mix. The download dialog shows the credit cost; AI downloads require an account and sufficient credits.
Try removing the crowd while keeping the scene
On the same original excerpt, try a concise description of the audible unwanted sound, such as “crowd applause” or “people cheering.” Select Isolate sound, then audition Background audio: that is the remaining mix after the target is separated. Compare it with the original and with the speaker-focused result.
Check applause that overlaps consonants, cheering over instruments, and the moments immediately before and after a reaction. Crowd voices can resemble the speaker; claps can overlap percussion; music and ambience may be reduced along with the target. If a candidate damages the scene, use a cleaner recording, a smaller local edit, or retain some crowd rather than treating separation as proof of recovery.
When a volume edit or noise reduction makes sense
For a shout confined to a pause, use clip gain or volume automation to lower that interval gradually. Leave the pause in place, preserve the speaker's breath and word boundaries, and check that the surrounding room tone does not suddenly disappear. If the shout overlaps speech, the same dip lowers the speech too. The local volume-edit guide explains how to shape a dip in Audacity.
A noise profile is useful for some steady sounds, but a changing audience is not a steady fan or hiss. Audacity's Noise Reduction documentation notes that variable sounds such as crowd noise are less suitable for that effect. A gate likewise responds to level rather than recognizing a crowd; it can cut quiet words and leave a loud cheer. Compare methods in the audio cleanup guide.
Return cleaned audio to the video
Import the chosen audio into your video editor at the same source start point and mute the original audio while comparing. If you processed an excerpt, align it with its original location and listen across both transitions; sudden changes in room tone can reveal the edit. Export a test and check sync at the beginning and end. The video audio replacement guide covers full-length and excerpt workflows.
Common questions
Can I remove a crowd but keep the music?
Try targeting the audible crowd and inspect Background audio, then compare the arrangement against the original. Shared frequencies and simultaneous claps or instruments can cause bleed or missing detail. Original tracks or a soundboard recording offer another source to compare.
Can I isolate one speaker from a crowd?
You can test a description of the desired voice, but similar or overlapping voices make the task difficult and a specific speaker is not guaranteed. Listen to the hardest overlap and keep an alternative recording where available.
Will removing crowd noise also fix echo?
Not necessarily. Reflections recorded with the speaker are different from crowd sounds and may remain. See the reverb guide before applying additional treatment.