What I mean is, not which app/option/button to use to remove a hissing sound, but how does the tool one uses to remove the hiss do that. Does it remove every low-volume noise? Does it break the sound into its components and remove a certain type of sound? etc.
Well, this depends strongly on the tool. In my opinion, the "easiest" (or at least the most intuitive) way to do it would be to use a filter that would remove the frequency content of the hissing sound. This could of course be done with the use of FFT-IFFT tools, but as already mentioned, this will strongly depend on the tool.
Additionally, it could make a difference if the "profile" of the hissing sound would be available to the tool or it would have to "guess" what a hissing sound is.
The best way to do it is to first examine a sample of the hiss. This means, you need to find a part where the hiss is the only sound that is being heard. If you have a denoising software, it will create the hiss profile, if not, you have to use a spectrum analyzer and manually check the frequencies that are increasing the most when the hiss is the only sound. Afterwards the denoising software applies a filter which is mostly an equalizer that cuts/reduces those frequencies. It can also be done Manually if the hiss is not too complex or doesn’t get in the way/conflict with many other sounds in the same frequency range. Of course this will also affect the sound as a whole. That’s why a clear source file is always the best.
A Filter/Equalizer which analyses the whole Spectrum, and higher Frequencies 2k- 10k a "hiss" can be removed by turning the "hiss" frequencies some db down.
In Vocal Production for example you have to find the "z", "s", "ch", "f" etc. frequencies as these are the "hissing" sounds, removing to much high frequencies will create chopped high frequencies which will result in u guessed it "hissing".
A noise sample based cleaning divides the sound file to chunks typically shorter than 0.1 seconds. Every chunk is converted to an equivalent Fourier spectrum. Every frequency component in the spectrum which is as weak or weaker than the same frequency component in the noise sample is silenced totally. The cleaned audio file is made by applying to every cleaned spectrum chunk inverse Fourier transform and saving the generated sound chunks to a new audio file.
The method is very effective against analog tape hiss and the noise of mics and mic preamps, because the noise is stationary => a single noise sample presents well the whole program. Audio recording & mixing programs have a button "capture noise sample".
Older noise reduction system simply turned off the whole audio if its level was lower than a selected treshold. Pro mixing desks have this in every channel and its name is "noise gate".
In analog sound recorders and movies was common a special volume dependent equalization. In recordings the easy to hear frequency midrange was boosted substantially if it was apparent that recording would increase the experience of noisiness. That increase happened if the signal level was weak. In loud places the noise is masked. The amount of the boost depended in a complex way of the signal level and midrange content. The In playback the process was reversed. Company named Dolby earned fortunes by holding the patents of the method.
Another somehow simpler, but still clever idea was ownend by DBX. It lifted weak signals louder in the recorder input and reversed the process in playback. Tapes which were recorded by using either Dolby or DBX coding were inferior if listened without the right reverse decoder.
A fairly comprehensive if not technical description can be found here: