A short clip, nothing to install, no upload: that is the case a browser handles well. The free Noise remover on this site runs inside the page, takes the video file itself, and saves the cleaned sound. You then put that sound back on the picture in your editor, or with one command.
This guide walks that route in four steps, then says when a clip is too long for it. Nothing is uploaded at any point, so there is no queue, no size cap from a server, and no copy of your video sitting somewhere else.
What the free browser route does
The tool runs the same AI model as the app, DeepFilterNet3, in a Web Worker inside the page.
- The file never leaves your computer.
- No account, no sign-up, no watermark.
- Files up to 10 minutes.
- It takes video files directly, not only audio.
The 10 minute limit is the price of that. Your browser has to hold the whole recording in memory, so a long file is turned away rather than half-processed.
Step 1: drop the video in
Open the Noise remover and drop the video on the page, or click to choose a file. It reads mp4, mov and webm, and every common audio format too.
You do not have to pull the sound out first. The page reads the video’s audio track and leaves the picture alone.
A file over 10 minutes comes back with a download button for the app. That is the point where you switch route, not a failure.
Step 2: choose how much to remove
There are three settings under Strength:
- All the noise it can. The default. Use it for fans, air conditioning, hiss and traffic.
- Medium, keeps some room sound. Use it when the room is part of the recording and you only want the worst of the noise gone.
- Light. Use it on a clip that already sounds close to right.
Press Remove noise and wait. A short clip takes a few seconds.
If the voice sounds thin or odd afterwards, run it again at Medium. Too much strength is the usual reason a cleaned voice sounds unnatural, and this guide explains why.
Step 3: save the cleaned sound
The page shows a before and after player. Switch between them while it plays and listen for what actually changed.
Press save and pick a format:
- mp3, smaller. Fine for a voice track that goes straight back into a video.
- wav, full quality. Use it if you will edit or level the sound again.
The file is named after the original with clean on the end, like clip-clean.wav. What comes out is audio. The tool does not write a new video file, so there is one step left.
Step 4: put the sound back on the video
In your editor. Drop the cleaned file on the timeline, mute or delete the old track, and line the new one up at zero. Then check one sync point near the start and one near the end. A lip flap or a hand clap is enough. The cleaned file keeps the original’s length and timing, so it should sit exactly where the old track sat.
From a terminal. ffmpeg copies the picture without re-encoding it and swaps the sound:
ffmpeg -i clip.mp4 -i clip-clean.wav -map 0:v:0 -map 1:a:0 \ -c:v copy -c:a aac -b:a 192k -shortest clip-final.mp4-map 0:v:0 takes the video stream from the first file, and -map 1:a:0 takes the cleaned audio from the second. -c:v copy copies the picture instead of re-encoding it, so it loses no quality. -c:a aac writes the audio in a format every player reads. -shortest ends the file where the shorter stream ends.
This command was run on a test clip while writing the page: the picture came out as h264, copied, with the cleaned sound in aac, and the length unchanged.
When the browser is the wrong tool
The clip is longer than 10 minutes
Use the app. It runs on your computer, takes recordings of any length, and is free for personal use up to 20 files a week with every tool. No account needed.
Drop the video straight in. Since version 2.0 every tool takes video files, and a video’s result is saved as m4a, so step 1 disappears: Remove noise reads the picture’s sound and writes the cleaned file itself. Version 2.2.0, the current one, also fixed videos with more than one sound track, where each step now reads the first track.
You have a batch of clips
The browser takes one file at a time. Remove noise in the app takes a list, so a folder’s worth of clips can be cleaned in one run.
You want the video back without a remux step
Then you need a tool that merges the cleaned track back for you. Auphonic does that: it processes the audio it extracts from a video and merges it back into the file. Its free tier is 2 hours of processed audio a month, and free productions carry a jingle at the end, which rules it out for anything you publish as it is.
The app route, in short
- Download Gleam Audio and drop the video on the window.
- Pick Remove noise and set Strength. Start at 100 and lower it if the voice sounds thin.
- Turn on Repair clicks only if the recording has pops.
- Save. A video’s result is saved as m4a, named with
clean. - Put it back on the picture with step 4 above.
If the sound is also uneven or too quiet, Enhance handles both in one pass. Pick the preset for the kind of recording, like Podcast, Lecture or class, or Room with echo, and it fills in the steps. Use Before and After on the result to hear what changed.
When it won’t help
- Music in the video. The model is trained on speech, so it takes music out along with the noise. On a clip where the music matters, use a lower strength or skip this step.
- A clipped take. Clipping cuts the top off the waveform and no cleanup puts it back. Fix the level before you record; the pre-upload checklist covers that side.
- A live stream or a call. Both routes work on finished files.
- Heavy echo. Noise removal takes out steady sound, not a room’s reverb. The app’s Room with echo preset tames some of it, but a bare room is a recording problem.
Takeaways
- For a clip under 10 minutes, the free browser tool is the fastest route. Nothing to install, nothing uploaded.
- It takes the video file directly and saves the cleaned sound as mp3 or wav.
- Put the sound back in your editor, or with one ffmpeg command.
- Longer than 10 minutes, or a batch: the app does it offline and saves a video’s result as m4a.
- Keep the strength as low as the noise allows. That is what keeps a voice sounding like a voice.