Cut the silences out of your video, in your browser

At a glance

• What it does: finds the silent passages in a video by loudness and removes them, with every cut shown on an editable timeline.
• Opens: MP4, MOV, WebM and MKV files, up to 2 GB.
• Exports: MP4 with H.264 video and AAC audio, at the original frame rate or 24, 30 or 60 fps, at the original resolution or 1080p or 720p (never upscaled).
• Editing: trim, split, merge, duplicate, reorder and delete clips, set the volume of each clip from 0 to 200%, undo and redo.
• Runs: in the browser, on your computer. Nothing is uploaded, there is no account and nothing to install.
• Price: free, with no watermark and no export quota.

The three settings that matter

Under the interface, silence removal is a loudness measurement. CutNoodle walks the audio track, marks every stretch that stays below a threshold for longer than a minimum duration, and rebuilds the video from what is left. The intelligence is not in the detection, it is in these three numbers and in the corrections you make afterwards.

• Threshold. How quiet counts as quiet. A room with an air conditioner has a much higher noise floor than a treated booth, and a threshold tuned for one will misread the other. If nothing gets cut, the threshold is too strict; if speech gets cut, it is too generous.
• Minimum length. How long a quiet passage has to be before it is worth removing. Set this to a fraction of a second and the result stutters, because normal speech is full of tiny gaps. Set it to three seconds and only the real dead air goes.
• Margin. How much quiet to leave on each side of a phrase. Cutting exactly at the threshold crossing clips the soft consonants at the start and end of sentences, so a small margin, usually a couple of tenths of a second, is what makes the result sound natural.

There is no universal setting. A lecture recorded on a lapel mic wants different numbers from a podcast recorded on a condenser, and you will find yours in about two attempts. How long should a pause be works through the reasoning if you want a starting point, and the full guide to cutting silence goes deeper into what the detector is measuring.

The three ways it gets it wrong

The first attempt is rarely the one you keep, and that is normal. The failure modes are easy to recognise, and each has an obvious fix.

Nothing was cut. Your noise floor is higher than the threshold, so the detector never hears a passage as silent. Rooms with fans, laptops with coil whine, and mics with a hot preamp all do this. Raise the sensitivity until the quiet stretches register.

Words got eaten. The threshold is treating quiet speech as silence, or the margin is too small. Trailing consonants and softly spoken sentence endings are the usual casualties. Lower the sensitivity, or widen the margin so more air is kept around each phrase.

The result stutters. The minimum pause length is too short. Ordinary speech contains dozens of gaps under a second, and cutting all of them leaves a machine-gun rhythm nobody wants to listen to. Push the minimum up until only genuine dead air qualifies.

This is why CutNoodle hands you a timeline instead of a finished file. A loudness detector has no idea what your video is about: it cannot tell a pause where you lost your train of thought from a pause you left before the punchline, and it cannot see the picture, so three seconds of holding up a product reads as silence. Playing through the joins takes a fraction of the time the manual edit would have, and it is where the two or three cuts that would have embarrassed you get caught.

What it will not do

CutNoodle cuts by loudness, not by meaning. It does not transcribe your audio, so it cannot remove filler words like um and you know, and it cannot give you subtitles. It processes one video at a time, with no batch mode and no timeline export to other editors. It is a single-track tool: no multicam, no colour grading, no titles, no music bed. And because the encoding happens on your machine, a long 4K file is real work for your laptop rather than something a rented server absorbs for you.

If you need transcript-driven editing, a tool built around speech recognition is the right answer: CutNoodle compared with Descript explains where that line falls. For the same cut inside a full editing suite see CutNoodle compared with Premiere Pro, for the command line version CutNoodle compared with auto-editor, and for the whole field at once all the alternatives.

Why it runs in the browser

Silence removal is a small, well-defined job that historically meant buying a desktop application or installing a command line tool. The other common answer is a web service, which solves the install problem by creating a worse one: your footage has to be uploaded to somebody else's computer before anything happens. On a two gigabyte recording the upload can take longer than the edit.

CutNoodle takes the third route. The encoder is ffmpeg compiled to WebAssembly, so the same engine desktop editors rely on runs inside this page. Your browser opens the file from disk, analyses the audio track only, which is why the scan finishes far quicker than a full decode, and writes the export locally. Nothing is transmitted, so there is no upload bar, no queue, no storage limit, and no account. Local video editor covers what that architecture does and does not buy you, and auto cut video silence covers when the automatic pass is worth running at all.

Common questions