▶TranscriptReel
Home / Tools / Silence Remover

Remove silence from audio

Find the dead air in a recording, look at every cut before you commit, then export a tightened WAV. Detection runs on the decoded samples, not on a picture of the waveform.

No file loaded. The audio track of a video file works too.

When cutting silence actually helps

There are two very different reasons to remove silence, and only one of them is about transcripts.

The first is for a listener. A meeting recording with a four-minute gap while people looked for a file wastes four minutes every time someone plays it back. Cutting the gap is straightforward editing and the result is simply a better file. It also makes the upload smaller, which matters on a slow connection.

The second is about the head of the file, and here we have to be careful, because the common claim is stronger than the evidence. We tested it on 2026-11-14: the base file was 50.86 seconds and returned 738 characters and 138 words. The same file with 8 seconds of digital silence spliced onto the front was 58.86 seconds and returned 737 characters and 137 words. That is a difference of one word, and the words were in the same order. So a silent head did not break the transcript for us, and the honest advice is to test your own file rather than assume.

What does not help is cutting every internal pause. Word boundaries are 50 to 200 milliseconds wide, and a tool set to a 100-millisecond minimum gap will chop a sentence into fragments, which makes the transcript read badly. That is why the minimum gap here defaults to half a second.

How is the 20-millisecond window measured?

The decoded buffer is walked in steps of 0.02 seconds × sample rate. For each window the page keeps the largest absolute sample, converts it to dBFS with 20 × log10(peak), and compares it to the threshold. A run of consecutive windows under the threshold counts as silence only if the run is longer than the minimum gap. Windows are never averaged, so a single loud sample in a window is enough to keep that window out of the cut.

Why did it find no silence in a file that sounds quiet?

Because the room is not silent, only quiet. If the noise floor sits at -32 dBFS and you set the threshold to -40, nothing qualifies. Raise the threshold until the detected sections match what you hear, then lower the minimum gap if it is merging pauses you wanted to keep.

What is the padding for?

Cutting exactly at the threshold clips the consonant right before the silence, which sounds abrupt. Keeping a tenth of a second on either side of each cut preserves the attack of the following word, at the cost of leaving a fraction of the silence in.

Is my file uploaded?

No. Everything happens in the tab. Nothing is sent to a server and nothing is stored.