If you work with voice agents, I think this is an interesting paper. The figure I would like to highlight is Figure 1. Basically, the distribution of times between turns in conversations is unimodal and the mode is somewhere between 0 and 200ms, in any of the 10 languages they show.