That's a great idea if the injected ads are exactly the same length and always get put in the same place. If you have a 30 second ad in one download and 40s in another, 10 seconds of legitimate podcast audio gets interpreted as an advertisement.
You "just" need to find all the common segments between different downloads; those are very likely to be the actual content, and not differing ads. Naturally this doesn't work very well if the pool of ads is low and you get repeats.