Some of the modules in iZotope RX might be worth trying out to eliminate some of the undesirable sounds and reveal the desired voices more clearly..
- You could try the Dialog Isolate module to try to remove everything except voices
- You could try using the Voice De-noise module to select & Learn the steady sounds of the AC & fan and remove them
- For more irregular unwanted sounds, you could try identifying & drawing them out graphically in the spectrogram display in the Spectral Repair module, though this can often involve a lot of trial & error
- If the voices are in the center of a stereo recording, you could try MS techniques to isolate/extract the center and reduce/eliminate background sounds on the left & right sides [the Center Extract module in RX also does this]
- The De-rustle & De-wind modules also address specific noise problems with dialog recordings [if either of those issues apply]
But of course if the dialog/whispers you're trying to isolate are very quiet and buried in the background noise, it may be much more difficult for the software to differentiate them, and if it does, to isolate them without the usual watery, chirpy artifacts characteristic of pushing this kind of the processing to its limits. But I'm sure it's worth a try -- you'll just have to experiment to see what might work..
This RX course shows how the various modules work: