I don’t have any definitive proof one way or the other, but I would be amazed if it was anything other than a voiceover.

He has a superb, natural feel to his commentary that doesn’t feel stunted and robotic like reading from a script often can be, bit it is also just too perfect to actually be a continuous take along with the footage - he never flubs a line or misses a cue, hits all of the main points of the process perfectly, and just nails it in a way that is too seamless.

The audio is also just a little too perfect - easy when sat before a mic in a recording setup, much harder to do consistently when moving around and working, switching between machines, and dealing with the foibles of your chosen mic system used live - no noise from the clothes you lav mic is attached to, or variable volume from moving nearer or farther from a desk mic…