Is a "preprocessor" model geared for a particular sound?

I’m confused about how preprocessor models work. My initial understanding was that a preprocessor was creating spectrograms of what it “hears” with little regard for the specific sound.

But in this file:

I read things like this:

_FILE_TO_TEST = flags.DEFINE_enum('file_to_test', 'no', ['no', 'yes'],
                                  'File to test')

To me, that implies the preprocessor is sensitive to specific keywords.

Can someone clarify?