I actually mean the YouTube videos; I haven't watched a stream while it was live yet. But, I believe it is in the actual input because on occasions where he had the recording program on the screen you can see it max out the peak volume meter (which corresponded to the clipping sound precisely).
For example, in video 020, when he says "hello" in the first few seconds, you can hear the clipping. At least it is clear as day on my headphones.
Here's a great video about setting input/gain level for voice recordings (which would apply for streaming just as well) :
https://www.youtube.com/watch?v=Sw8ppTPUXlo