Discontinuous transmission stops sending audio packets when the talker is silent, instead of wasting bandwidth encoding the sound of a quiet room. Driven by VAD, it can cut the data a typical conversation sends by 40-70%, since people are silent more than half the time in a two-way call. To keep the line from sounding dead, the encoder periodically sends a tiny descriptor of the background noise (a SID frame) so the receiver can generate matching comfort noise during the gap. DTX is especially valuable at scale — large conferences, mobile networks, paying for egress — but it must be tuned so onsets aren't clipped and the noise fill sounds natural.
