I am very confused about a behaviour I can’t explain.
When I normalize a track to a certain peak amplitude (e.g. 0 DB) and then analyze the track with “plot spectrum”, I would expect to have maximum peak at 0DB. Even when I choose size=65536, I still get a peak which is a LOT lower (e.g. -21DB).
What am I missing here?
“Plot Spectrum” measures the sound, spread across multiple frequency ranges (“frequency bins”). The amplitude measured by Plot Spectrum, will be about the same as the peak amplitude, if the sound is all in one “frequency bin”. If the sound falls into multiple frequency bins (as is the case with most natural sounds), then each bin will be less than the overall peak level.
Ok, but still when I look at the waveform of the track, after normalization the highest blue line doesn’t reach the top/bottom of the area, there’s still quite a gap to the border of the area. When I look at commercial music tracks, the waveform often reaches the top/bottom of the area.
The only explanation I would have for this is, that there are very short ranges in my audio which are much louder than the average but so short that they don’t get drawn in the wave form and also not “filling up” one frequency bin of the spectrum plotting. But that seems quite unlikely.
We are warned against jumping straight into the weeds of comparing dB values without finding if there is an actual job back there somewhere. @edelm probably didn’t wake up one day saying, “Today, I think I’ll compare spectrum values.”
Much more likely there is a job such as wanting to make a sackbut recording as loud as possible without clipping damage, as an example, and not being able to resolve Audacity analyze tool results.
The Audacity limiter is very good (it can limit without clipping/distortion). Use make-up gain or Amplify or Normalize after to bring-up the overall volume.
Compression can help too, again with make-up gain. Dynamic compression reduces the dynamic range by making loud parts quieter and/or quiet parts louder. In practice, usually the loud parts are “pushed down” and then make-up gain is used to bring-up the loudness.
Limiting is a kind of fast-compression. Compression is harder to use than limiting because there are more settings to mess with.
Automatic volume control or automatic leveling is another kind of (slow) compression.
All of this is done by amplitued without regard to the spectrum.
Professional mastering engineers sometimes use multiband compression (which does compress different frequency bands separately). Audacity does not have this effect built-in.
Note that too much compressinon (or limiting) can make music boring (no dynamics = all one constant-volume), you can get side-effects, and sometimes it can sound like distortion.
With amplifying to 0.1 and show clipping I can identify the loudest spots quite good (would be nice to have that automated).
I could increase the loudness of the track quite a bit. Still not as “loud” as I want to, compared to commercial tracks, though.
Most of the time I find one track which I can reduce the volume/equalize, but then the next one is limiting the overall loudness of the track, so it’s a constant process of mixing/equalizing and then checking again.
I work a lot with compressors, but don’t dare to use a hard limiter.