If you want to compare two spectra (let´s call them A and B in this example), one way of doing it is to export the numerical values from Audacity into e.g. Excel. Then you divide A by B value for value, and plot the result as a function of frequency. Where A and B are different, the resulting curve will differ from one, giving you an impression of the relative difference between A and B. 

Personally, I would not use Audacity for this. I would use a scientific Pyhon distribution such as Anaconda or Python(x,y), and a Short Time Fourier Transform (FFT is not time invariant) for calculation of the spectra.