Trailer auto-generation?

Question

Video processing

Let's take Die Hard as an example.
Purpose: to programmatically generate a 2-3 minute trailer or even rather a video, preview content.
How would you do it?
Are there examples or what to read on this topic?

That's how I would do it, I just took moments 3-15 seconds long where the absence of sound began and the end is the end of the sound. For example, it can be a piece of dialogue or some kind of batch, action.
Meaningfully, I think it’s unlikely that it will come out to tie them, and it’s not worth it because of the spoilers, take such segments of 10-20 pieces and, in theory, it’s ready, right?

Answer 1

Or maybe ± 10 seconds from the "splash" of the sound level is better? And then, I'm afraid, between the absence of sound and the beginning we get silent scenes.
If the software implementation is of interest, then ffmpeg comes to mind, we build a histogram with it, we find local maxima on it.