I am comparing two sound signals in the frequency domain. As they can have been recorded in different volumes I need to normalize them. My initial approach was to divide each sample by the mean absolute value of all samples. I done this in the frequency domain (after a fft).
Now I realize that this can be the wrong approach, I could normalize in the time domain or I even could use the average power (amplitude squared). Each approach could yield different results. Of all those possibilities I still don't know which is more suitable to my case.