You don't need AI for that. 20 years ago my neighbors had a small device that would automatically mute the TV whenever the ads started playing and unmuted when they finished. No idea how it worked but for sure it didn't have a fancy neural network inside.
That device sensed the higher average sound levels at which ads are usually broadcasted, and it's also easy to build. The problem is those ads that have or start with lower sound levels which would be impossible to spot without matching them with a known audio/video pattern.
