Silence

Dead air and silence removal, without the edit being heard

A forty-minute interview is rarely forty minutes of talking. Taking the empty parts out is the fastest edit there is, and the easiest one to do badly.

What is being looked for

Silence in a recording is not silence. It is the room - the fan, the traffic outside, the hum of the desk - a few decibels above nothing. Removal works by finding stretches that stay under a level for longer than a length, and shortening them.

Three numbers decide everything.

Threshold. The level under which audio counts as a gap. Measured against the recording, not against zero.
Minimum length. How long a quiet stretch must last before it is one. Below about half a second you are cutting the pauses inside sentences.
What is left behind. Shorten to, rather than remove.

Shorten, do not delete

The gaps are shortened to a length that still sounds like a pause. Removing them entirely is what makes an edit audible - speech without breath is speech that has obviously been cut.

Speech with the gaps taken out entirely sounds like a machine reading a list. Leaving three or four tenths of a second keeps the rhythm of somebody thinking, and the edit stops being audible. A pause before an answer is often the most useful second in an interview.

Why a threshold on its own chops words

A word does not start at full volume. Set the threshold too high and the shaded parts are treated as silence, which is why badly cut speech sounds clipped at both ends.

The fix is not a lower threshold on its own. It is a short guard either side, so the gap begins a moment after the speech stops and ends a moment before it starts again. Seston keeps that guard, which is why a tightened take still has its consonants.

Set the threshold from the room, not from a number

A good threshold sits above the room and below the quietest speech, and where that is depends on the recording. A kitchen at −48 dBFS and a treated booth at −64 dBFS do not take the same setting.

Seston measures the noise floor before it does anything, so the level it proposes comes from the file rather than from a default. If the recording is so noisy that the floor and the speech overlap, the answer is not a cleverer threshold - it is removing the noise first and then cutting.

Do this before you level, not after

Loudness is measured across the whole programme, so a take full of dead air measures quieter than it sounds. Cut the gaps first, then normalise, or the number you hit will be wrong by however much silence you were carrying. More on that here.

What Seston does

Smart Edit finds the gaps, shortens them, and says what it found before it touches anything - how many, how much shorter the take becomes, and where. Nothing is written to the recording: clips point at the file and the edit lives in the session, so the take can be put back at any point.

It works on the selection or on the whole clip, and the two settings that matter are on the panel rather than buried in preferences.

Seston is free for Windows.