Dead air comes out. No cut ever lands inside a word

The input is one talking take, filmed on a phone, up to three minutes.

On the way to a finished ad, the machine transcribes every word with timings, finds the silences, and tightens the ones that run long. The same pass, at the fast pace, drops the hesitation sounds the transcriber flagged.

What comes out the other end is a 9:16 MP4. Cut, captioned, covered with footage on the words you marked.

That is the honest scope. There is no mode that strips silence from a file and returns that file. If you want your podcast episode back minus the pauses, a no-cost tool already does it: Audacity has shipped a truncate-silence effect for years, and most phone editors have a version of the same thing.

So this page is for one reader. Somebody making short vertical ads from their own spoken take, who wants the dead air handled without ever touching a waveform.

There is no silence slider. You pick a pace, and the pace sets the threshold.

At chill, only a pause longer than 700 milliseconds is touched. At normal the line is 450. At fast it is 240, short enough to bite into the breath between fluent sentences.

A tightened pause is not deleted. It is shortened to a breath: 300 milliseconds at chill, 200 at normal, 130 at fast.

And a pause at the end of a sentence keeps 60 milliseconds more than one in the middle of a clause. A beat between sentences reads as intent. The same gap mid-clause reads as a lost train of thought.

That is why the output does not sound gated. Every silence a listener expects is still there. It is just no longer three times longer than it needed to be.

Most silence removers are volume gates. They watch the waveform, and when the level drops below a line for long enough, they cut.

Speech does not stop at word edges. Consonants trail off, and the s at the end of a word can sit below a gate's line while it is still being said. A gate clips it, and the word arrives amputated.

The breath is worse. People inhale before a sentence, and a gate reads an inhale as noise to remove. Strip every intake and the voice never seems to breathe, which the ear flags as wrong before the brain can say why.

A sentence end is different. The pitch falls, the word completes, the breath resets. A cut there matches what the listener already expected, so it registers as delivery rather than as an edit.

So the rule here is absolute. Silences are shortened, whole lines can be deleted, and the flagged ums go at the fast pace. Nothing is ever cut halfway through a word.

The tightening is automatic, and automatic means occasionally wrong. A dramatic pause you planted on purpose is, to a threshold, indistinguishable from a lost thought.

The first fix is the pace setting. Chill leaves everything under 700 milliseconds alone, which covers almost any pause a person places deliberately.

The rest of the cleanup is yours, on the transcript. Delete the line where you restarted and the cut rebuilds around the gap. Change the pace and the whole read re-tightens to the new threshold.

Your original take is never touched. Every version is recomputed from the file you uploaded, so trying the fast pace and stepping back to normal costs nothing before export.

One trip through the machine is 100 credits: transcription, the silence work, the directing pass and the footage search. Export is 20 credits per minute of finished video, so a thirty second ad is about 110 credits all in.

Three finished videos are free inside your first seven days, no card, on the 600 credits granted at signup, and they export with a Cutroom mark across the middle. Lite is $19.99 a month for 1,250 credits, nothing on the picture of a paid export, cancel in one click.

If your file is a podcast, a webinar or an hour of anything, this is the wrong shape twice over. The uploader stops at three minutes, and the output is always a finished vertical ad, never your cleaned original.

For that job, take the no-cost route: Audacity for audio, your phone's own editor for video. They hand your file back, which is exactly what this does not do.