SubForge Captions

Remove silence from video, word-accurate

Cut the pauses and the “um”s out of a talking-head video. You see every cut before it happens, and get subtitles that match. Nothing is uploaded.

How to remove silence from a video

  1. Drop your video. MP4, MOV or WebM, up to 10 minutes. Or press Try a sample clip.
  2. Press Find pauses and fillers. The speech model writes down every word with its timing. The first run downloads the model once; after that it starts at once.
  3. Choose what goes. Every pause longer than your setting, and every filler, shows on the timeline. Click a block to keep it, drag the pause slider, and play the preview: it skips the cuts as it plays.
  4. Export. Press Export cut video for an MP4, and download SRT or VTT subtitles that already match the new timing.

How it finds pauses and fillers

Most silence removers look only at loudness, so they cut quiet words and keep loud breaths. This one starts from a transcript: Whisper large-v3 turbo, on your graphics card, writes every word with its start and end time. A pause is the gap between two words, and its edges are checked against the sound, because the model often stretches a word's end into the pause after it. Fillers are words from a list for your language: “um” and “uh” anywhere, and words that are also real words (“like”, “you know”, “I mean”, “o sea”, “tipo”, “du coup”, “يعني”, “मतलब”) only when the transcript sets them off with commas, so “I like it” stays.

Voice inside a pause with no word attached is shown as its own block: usually a breath, a cough, or an “um” the model chose not to write. Speech models are trained on clean transcripts and tend to leave disfluencies out, so on some recordings this catches more fillers than the word list does. It is off by default because it can also catch a breath you want to keep.

Jump cuts, smoothed

Cutting a pause out of the middle of a talking-head shot makes the face jump. Tick Smooth jump cuts and every other section is punched in by 8%, the trick editors use to make a jump cut look deliberate, and the audio joins get a softer 40 ms fade. Without it, joins still get a 12 ms fade so they never click.

Subtitles that match the cut

The words are already timed, so the subtitles for the shorter video come for free: every word moves with its section, cut words (the fillers) are dropped, and the lines are rebuilt to two rows of 42 characters (26 for vertical video). Download SRT for YouTube, Premiere, Resolve and CapCut, or VTT for web players, and translate them with the subtitle translator.

Settings that work

VideoCut pauses overKeep around speech
YouTube talking head, fast pace0.4 s0.10 s
Lesson or tutorial0.6 s (default)0.15 s
Interview or podcast video1.0 s0.25 s

More free video tools

Privacy

Your video is read, transcribed and cut by your browser on this page. It is not uploaded, stored or seen by anyone. The only download is the speech model, once; your browser keeps it for next time.

Questions

How do I remove silence from a video for free?

Drop the video into the tool on this page and press Find pauses and fillers. Every pause longer than your setting is marked; press Export cut video to download the shorter MP4. No account, no watermark.

How is this different from a silence detector?

A silence detector cuts by loudness alone, so it can clip a quiet word or keep a loud breath. This tool works from the words: it knows where each word starts and ends, cuts only the gaps between them, and checks each gap's edges against the sound.

Can I keep some pauses?

Yes. Click any pause on the timeline, or untick it in the list, to keep it. Raise the pause length to cut only the long ones, and use Keep around speech so cuts don't feel rushed.

Does it remove filler words too?

Yes: “um”, “uh”, and “like” or “you know” when they are set off by commas, in seven languages. See remove filler words.

Will the subtitles still line up?

Yes. The tool re-times every word to the cut and rebuilds the lines, so the SRT or VTT you download matches the exported video.

What does Smooth jump cuts do?

It alternates the framing between full and a slight 8% zoom at each join, so a cut in the middle of a talking-head shot looks like a deliberate edit, and it softens the audio joins.

Is my video uploaded?

No. Transcription, the cut and the export all happen in your browser. The file never leaves your device.

Is it free? Is there a watermark?

Free, with no watermark and no sign-up, for videos up to 10 minutes. Longer files will be part of SubForge Pro.

Which browsers work?

Recent Chrome and Edge on desktop are best: they have WebGPU for fast transcription and WebCodecs for the export. Safari 26 works too. Firefox can transcribe but may not be able to export MP4 yet. Without WebGPU the transcription runs on the CPU, several times slower.

What languages work?

Pauses work in any language Whisper can transcribe (about 99). Filler lists are built in for English, Spanish, Portuguese, French, Hindi, Urdu and Arabic, and you can edit them.

Credits and licences