Most production voice stacks are three systems stitched together. One model transcribes, a second separates speakers, and a detector decides when the user stopped ...