There are some hardware devices such as various AGC (automatic-gain control) components that can do moderate leveling, but since they have to operate in real time (., without look-ahead ), they can't do much. And they aren't cheap, let alone free. Even a skilled human can only react to changes unless s/he is lucky enough to be present during a recording session and can use visual cues to anticipate coming variations. Software can do better by performing multiple passes over the audio, generating a loudness map of where the volume changes. (It's not actually that simple, but the metaphor is helpful.)
A good conversation lesson can often begin with the students not being sure what the topic of the day is. Not that we want to keep them in the dark for long, but it can get their attention when things aren’t so clear from the outset. But when I teach this lesson I introduce the topic directly, simply to give me a chance to point out that nobody will be required to discuss personal matters related to money. We don’t need to know how much money you’ve got or earn, I may tell them. It can be a relief, on occasions, to those who do consider this a very private matter.