I don't think you want to use Simon Bosley's solution for this situation because if you do, you will be left with a mono track with any similar information attenuated or completely missing. As I don't know what the recording is like, this could be a lot of information. e.g. drums, low frequencies, orchestra, main vocals, etc.
I would instead use a more subtle approach. I would use a Mid/Side EQ to attenuate the specific frequencies from the mid channel. This will leave you with a stereo file with most mono information in tact.
If you are capable, you could refine this more. You could isolate the most active frequencies from the conversation, and use it as a "side-chain" to control the wet/dry mix of the mid/side EQ. You would probably need an envelope follower or a gate or something with side-chain capabilities, and you would need to do some routing.
If Audacity doesn't allow you to do the second bit, or if it's a bit tricky, you could manually record and edit automation of the wet/dry mix. Assuming you can automate parameters in Audacity.