Take the sound out of a video and you may want two very different things. Either the picture should stay and the sound should go, so you have a silent clip. Or the sound should stay and the noise inside it should go, so the voice is clean. The steps for one do nothing for the other, and most guides answer only the first.
This page does both, in the order you would use them. Gleam Audio covers the second job and runs offline on macOS, Windows and Linux. The first job is a video editor or one command, because the app returns audio and never writes a video file back.
First, decide which job you have
| What you want | What happens to the sound | What does it |
|---|---|---|
| A silent video | removed, gone | your editor, or one ffmpeg command |
| A clean video | noise removed, voice kept | Gleam Audio, then the audio goes back |
If you are not sure, ask what you would do with the result. A silent clip is for a background, a timelapse or a clip you will dub later. A clean clip is for a video where someone talks and the room was not quiet.
1. Make the video silent in your editor
Every editor can do this, and the shape is the same in all of them. Select the clip on the timeline, mute it, or detach its audio and delete the detached piece. Then export as you normally would.
Two things are worth knowing before you take this route.
- The export re-encodes the picture. Most editors encode the video again on the way out, so a mute costs one generation of picture quality. That is fine if you are still cutting. If the picture is finished, the command in the next section copies it instead.
- Mute the clip, not the track. Muting a whole track also silences anything else you put on it, including music you meant to keep.
2. Make the video silent with ffmpeg
If the picture is done and you only need the sound gone, ffmpeg copies the video stream and drops the audio track, so nothing is re-encoded:
ffmpeg -i in.mp4 -c copy -an silent.mp4-an means no audio. -c copy means copy the streams rather than re-encode them, which is why the file shrinks by roughly the size of the audio track and the picture is untouched. Run against a test clip on ffmpeg 9.0.1, the result holds one stream, video, and no audio stream at all.
The same line works on a mov or an mkv. Keep the extension the destination expects, since the container changes what some players will open.
3. Remove the noise from the sound, and keep the audio
This is the job Gleam Audio does, and since 2.0 it does it on the video file itself. Every tool takes mp4, mov, mkv, webm and avi for their sound. The result is audio only, saved as m4a, or in the format you pick in Convert.
- Drop the video on the window, or click Open Files…. Adding a video adds its sound, not a copy of the picture.
- Pick Remove noise in the sidebar. It is tuned for speech, which is what most videos carry.
- Set Strength. Start low, and raise it only until the quiet gaps between words are clean. Too much is what makes a cleaned voice sound robotic.
- Choose where the result goes in Save to, and set the name pattern in Name. The default is
{name}-{tool}, sotalk.mp4becomestalk-clean.m4a. - Click Run at the top right.
Version 2.1 adds a Before and After switch next to the play button of a finished file. Switch while it plays and it jumps to the same moment in the other file, so you can hear what the strength setting did without hunting for the same second twice.
If the video needs leveling too, Enhance works on the same file in the same run list. Loudness is a separate decision from noise, and what LUFS means is the number both YouTube and podcast hosts measure you against.
4. Put the cleaned audio back into the video
Now the video needs its sound back. In an editor, drop the cleaned file on the timeline, mute the original clip’s audio and export. That works, and it is the route to take when the picture is still being cut.
When the picture is finished, put the track back without re-encoding it:
ffmpeg -i in.mp4 -i clean.wav -map 0:v -map 1:a -c:v copy -c:a aac -shortest out.mp4-map 0:v takes the picture from the first file, -map 1:a takes the audio from the second, -c:v copy leaves the picture alone and -c:a aac writes the new track in the codec mp4 players expect. -shortest ends the file at whichever is shorter, so a track a fraction of a second off cannot leave a silent tail. Checked the same way: the result holds an h264 video stream and an aac audio stream.
Two things decide whether this step goes smoothly.
- Length. Remove noise, Enhance and Convert keep the length and the timing, so the new track lines up by itself. Remove silence and Change speed do not: they shorten the file, and a shorter track drifts away from the picture within seconds. Do those only when the picture goes with the sound, or not at all.
- Channel count. A mono track on a stereo video plays out of one side on some players. Convert can write the channel layout you need, or set it when you record.
Watch one sync point after the swap, a hand clap or a door, and confirm the picture and the sound still agree.
5. Get the audio out on its own first
Some jobs want the audio as a separate file before anything else touches it: a transcript, a second edit, or a client who asked for the sound. Two ways, and neither uploads anything:
ffmpeg -i in.mp4 -vn -c:a pcm_s16le audio.wav-vn drops the picture, and pcm_s16le writes plain WAV, which is the cleanest thing to hand an audio tool. On a test clip that produced a 44.1 kHz mono WAV, matching the source.
Or use the mp4 to mp3 tool for a single short clip, which saves an mp3 straight from the browser.
When it won’t help
- Removing one voice from a mixed track. If two people talk over one microphone, or a voice sits on top of music, no cleanup tool separates them. That is a re-record or a different take, not a setting.
- Removing music and keeping the speech. The noise remover is trained on speech and treats music as something to take out. On a video with a music bed under a voice, it will thin the music before it thins the room.
- Putting back what a compressed track lost. A phone video records compressed audio, and the detail is already gone. Cleaning makes the hiss quieter. It cannot restore a voice that was never captured well.
- A video you only have as a link. Every step here needs the file. Download it first, at the quality you intend to publish.
- A silent video that has to stay the same size. Dropping the audio track makes the file smaller. If a platform or a client wants the exact file back, mute in the editor instead.
If the problem is noise rather than the sound itself, the guide to reducing background noise in a video walks the same job from the other end, and the complete guide to cleaning up spoken audio puts the order of work together. That earlier guide was written before the app took video files, so it exports the track first; the app now opens the video directly.
Takeaways
- Two jobs, two answers. A silent video comes from an editor or
-c copy -an. A clean video comes from cleaning the audio and putting it back. - ffmpeg copies the picture, so a mute or a track swap costs no picture quality. An editor’s export usually does.
- Gleam Audio takes the video directly since 2.0: mp4, mov, mkv, webm and avi, with audio out as m4a.
- Keep the length. Remove noise, Enhance and Convert preserve timing. Remove silence and Change speed do not.
- Check one sync point after the track goes back, before the file leaves your machine.