Remove the silences from a recording
Two hours of lecture with half an hour of speech inside. Here all the pauses are found on their own, you see them highlighted on the waveform before you touch anything, and the file comes out shorter, with the minutes you saved written at the top. The recording stays on your device.
Turning silence down and taking it out are two different things
This is the mix-up that wastes the most time, and it is worth writing out in full. A gate, like the one inside Enhance audio & voice, turns the noise down when nobody is speaking: the file stays exactly as long, the only change is that you no longer hear the fan during the pauses. Here the pauses are taken out: what comes out is a shorter recording, and the big number at the top tells you by how much.
They are two jobs that go one after the other, not one instead of the other: first you clean the noise up, then you shorten it. If the recording is dirty it is worth going through Enhance audio & voice and coming back here afterwards, because on a clean background the threshold finds the pauses far better.
The threshold is relative to your voice, not a fixed number
A fixed value in decibels cannot work, and the reason is simple: a memo recorded half a metre from the phone and one recorded with the phone in a pocket are twenty decibels apart. Here the level of your speech is measured on that file with the double gate of the ITU-R BS.1770 standard (the same one Your video's loudness uses for LUFS), and the slider says how far below that voice things start counting as silence. The value you find already set sits halfway between your voice and the quietest point of the recording.
The margin is on by default, a hundred and fifty milliseconds each side, and it is not a detail: without it the threshold eats the breaths and the unvoiced consonants at the start of a word, and that is exactly how these tools end up handing back speech that sounds torn. If you hear words being clipped short, raise the margin before you touch the threshold.
When it does not work, it says so
On a recording with the fan running, with the street open or with the microphone too far away, silence is not silent: between the voice and the background there are ten decibels instead of forty, and any automatic threshold gets it wrong. In that case what you find here is how many decibels apart they are rather than a result that merely looks good, and the choice comes back to you. The same goes for music, where pauses do not exist and there is nothing to take out.
And if with the chosen settings there is no pause long enough, nothing gets cut at random: the page says so. That is why the waveform is there before you download, with the stretches that will be removed in grey: they are there so you can notice that the threshold is eating words, instead of finding out by listening to the file afterwards.
What it does not do
It does not cut a stretch you choose: if you need to remove one piece, picked by hand, that is Cut audio. It does not speed anything up: Change speed & volume makes everything run faster, silences included, and changes the timbre of your voice, whereas here the words stay at the speed you said them. And it does not merely measure: if the question is «how loud am I compared with everybody else», the answer is in Your video's loudness.
If you load a video only its audio track is used, and what you download is audio: the video is not put back together. To pull the audio out and keep it as it is there is Extract audio from a video, and to really shorten a video there is Trim a video. The output is a 48 kHz WAV, that is uncompressed: it weighs a fair amount, and to put it back into MP3 the step is Convert an audio file.