Silence detection
Also known as: gap detection, dead air detection
Finding the stretches of a recording with nothing meaningful on the audio track, which is how dead time gets located cheaply.
Updated
What it means
Silence detection locates the quiet. True digital silence is rare on a camera, so in practice it means finding stretches whose level stays below a threshold for long enough to count as nothing happening.
It is one of the cheapest signals available about a recording, needing no model at all, and it correlates usefully with Dead time in the picture. A clip with no sound is often a clip with nothing in it.
The reverse inference is weaker. Plenty of interesting footage is near silent, which is why silence is treated as a hint to look closer rather than as grounds for discarding anything.
How it works in practice
The threshold has to sit above the noise floor rather than at zero. A camera in a quiet room still records a hiss, so the measurement is relative to that floor instead of to absolute silence.
Duration is as important as level. A gap of a fifth of a second is a pause between words, while twenty seconds below the threshold is a camera somebody forgot to stop, and only the second is worth acting on.
Combining it with speech boundaries makes it genuinely useful, because Speech-to-text tells the difference between a silence inside a sentence and one after it.
What to watch out for
Wind guarantees false negatives. An exposed microphone records constant noise well above any sensible threshold, so a whole ride can read as continuously busy with nothing on it.
Music laid under the footage breaks it entirely, which is why the measurement belongs on the original camera track rather than on an assembled Soundtrack.
Treating silence as grounds for deletion also removes good material. A long still landscape or a quiet moment before a Drop-in carries no sound at all and is frequently the shot a sequence needed most for contrast.
How RawClip handles it
Stretches with nothing on the camera audio are weighted down rather than discarded, because quiet footage is sometimes exactly what a sequence needs between two loud moments. The measurement runs on the original track, before any music is added.