I have the same issue with late night talk shows / live comedy shows where the audience cheering is way louder than the host's voice. At some point I tried an AI model to detect and lower the volume of these parts, but didn't get that great detection with the limited training dataset that I labeled manually. IMO Youtube or other streaming platforms should offer this option.