YouTube uses a Content Identification tool that enables copyright holders to find instances of copyrighted material so that they can then issue a takedown notice, “moneytize” it, or track it. This includes identifying copyrighted material on a video’s audio track and automatically muting it. There’s no mention in Google’s discussion of Fair Use exemptions.
Mark Smitelli has poked around at the system, uploading copies of the copyrighted song “I Know What Boys Like,” sonically altered in various ways: compressing or expanding the time, lowering or raising the pitch, adding noise, etc. Mark runs the complete results, but to roughly summarize: Altering parameters more than 5% often seems to fool the Identifier, and using less than 30 seconds also seems to let the clip slip through the rule-bound robot’s shiny little nets. Playing clips in reverse confused the Identifier, but stripping out everything except the vocals did not.
Using a clip for as satire [later: I probably got this wrong] or political commentary undoubtedly wouldn’t keep it from the Identifier’s snares, although such use is likely protected and non-infringing. The Identifier, unsurprisingly, seems to be a poor reader of human intention. [Thanks to David Abrams for the tip.]