Gain, measured rather than guessed
Changing volume digitally means multiplying every sample by a constant. That part is trivial. The interesting question is which constant, and the answer depends on how loud the file already is — which is why this tool measures your file's peak level while decoding it and shows you the number before you touch anything.
A recording peaking at 30% of full scale has plenty of room: you can more than triple it. One already peaking at 95% has almost none. The same slider position means completely different things in those two cases, and knowing the peak turns the adjustment from a guess into a decision.
Decibels and why the scale feels odd
The readout is in decibels because human hearing is logarithmic. Doubling a signal's amplitude is +6 dB, and halving it is −6 dB — the same perceived step in both directions, which a linear multiplier does not give you.
A useful reference: 3 dB is roughly the smallest change most people notice on music, 6 dB is clearly audible, and 10 dB is commonly described as about twice as loud. If a recording sounds slightly quiet, you are usually looking for 3 to 6 dB rather than doubling.
Clipping
Samples are stored in a fixed range, and there is nothing above the top of it. When gain pushes a sample past full scale the value has to be truncated, flattening the top of the waveform. That flattening introduces harmonics that were never in the recording, heard as a harsh crackle or fuzz — worst on transients such as drum hits and consonants.
Clipping is permanent. The samples that were cut off cannot be recovered by turning the file back down afterwards, which is why the warning appears before processing rather than after.
Peak normalisation and its limits
Normalising computes the exact gain that puts the loudest sample just below full scale, and applies it. For a quiet recording that needs to be usable, this is the correct one-click answer.
What it will not do is make several files sound equally loud. Peak level and perceived loudness are only loosely related — a heavily-compressed pop track and a sparse acoustic recording can share an identical peak while one sounds obviously louder. Matching perceived loudness requires measuring in LUFS and is a different operation.
Fades
A fade multiplies by a ramp instead of a constant: from zero up to full over the fade-in, and back down over the fade-out. Even half a second at each end removes the click that an abrupt start or stop produces when a waveform is cut mid-cycle.
Longer fades are an editorial choice — two or three seconds reads as a deliberate ending on music, while anything over about five starts to feel like the track is drifting away.