An average spectrum answers questions a waveform cannot: where the energy actually sits, whether there is rumble under the music, and what a codec threw away.
Reading the plot
Frequency runs left to right, level bottom to top in dBFS. The filled curve is the average across the whole file. The fainter line above it is the peak hold.
A full-scale sine reads 0 dB at its own bin, so the numbers are comparable between files rather than being relative to whatever is loudest.
FFT size
Each window is Hann weighted and overlapped by half, and the results are averaged. Size sets the resolution of a single window.
1024 samples resolves 47 Hz per bin at 48 kHz, which cannot tell a low E on a bass from the note above it. 8192 resolves 5.9 Hz, which can. The cost is that a 170 ms window averages over anything faster than that.
Log and linear
The log axis gives every octave equal width, which is how music is spaced. 20 to 40 Hz takes the same room as 10 to 20 kHz.
The linear axis is the one to use when looking for a codec cutoff or a single whistle, because the top of the band is where most of the width goes.
What to look for
Rumble below 30 Hz. Traffic, air conditioning, and handling noise land here. It is usually inaudible on small speakers and still eats headroom on every one.
A cliff at 15 to 16 kHz. The file passed through a low-bitrate lossy encoder at some point, whatever it says now.
A narrow spike at 50 or 60 Hz. Mains hum. The hum remover takes it out without touching the material either side.
A wide gap between average and peak hold in the low mids. A resonance that only rings on certain notes, usually a room mode or an untreated corner.
Limits
This is a whole-file average, not a spectrogram. It cannot show that the bass arrived at 1:20 and left at 2:40, only that it is in the file.
It also measures what is there, not whether it is right. A spectrum that looks unusual against a reference track is a place to listen, not a fault on its own.