First of all, what you want is known as "Source Separation" in research circles, and it has been mathematically proven to be impossible, in the general case. I.E., sometimes the "show" and the "noise" are too similar.
Now, the good news: Much progress has been achieved in the last 2 decades or so. See this Audition Tutorial:
Adobe invented the above Photoshop-like erasure technique, which has been promptly cloned by other applications, like iZotope RX4:
See the most recent bandwagon jumper here:
For the techies out there, this is the explanation of how it works:
"Imitation is the sincerest form of flattery"