That's true, but what they're talking about is comfort noise generation. There's always static on the line (background noise in the room for example and electronic noise) and as part of the compression, if the sound power is too low, no audio is sent. That's called silence suppression, and prevents the consumption of bandwidth when no one's talking (which is more than 50% of the time ... normally people aren't talking both at once).

Well, on the other end, during a silence period, nothing at all would be played, so it would sound like a dead line. Comfort noise generation does a bunch of math on the background noise at the transmitting end to pick up key frequencies in the background noise, and then these are recreated at the other end. They don't match (not even close -- you could consider it extremely lossy compression) but it's close enough to our ears so it sounds continuous.