Dead air is the bulk of most raw recordings: the pause before an answer, the gap while someone reads a note, the run-up before the first word. Removing it by hand is the slowest part of an edit.
Threshold
Threshold is the level a span has to stay under to count as silence. It is not zero, because no recording reaches zero: a microphone always picks up the room, the preamp, and the air.
The spans that will be cut are drawn on the waveform as you move the control. That is the fastest way to set it: raise the threshold until the marks cover the gaps, and stop before they reach into the words.
Shortest gap
Gaps shorter than this are left alone. It is what protects the natural pauses inside speech, which are usually 100 to 400 ms.
700 ms is a good starting point for an interview. Raise it towards 1.5 s to remove only the long stalls, or drop it to 300 ms for a tight promo edit.
Keep
Cutting every pause to nothing produces speech with no breath in it, which is exhausting to listen to. Keep leaves a fixed amount of silence at each edit, split evenly between the two sides.
150 ms is close to the pause a speaker takes naturally between sentences.
Blend joins
Each kept span is faded in and out over a few milliseconds at the edit. Without it, cutting from a quiet gap into a loud word steps the waveform, and that step is a click.
What it is not
This is a gate on the timeline, not a gate on the level. It removes spans of time rather than pushing quiet passages down, so nothing about the surviving audio changes.
To reduce the level of background noise while keeping the timing, use the noise gate. To keep the pauses but even out the volume, use the compressor.