In this lesson
How to read a spectrum analyzer
FFT size, window and averaging decide what an analyzer can show. Learn to set it for the question you are asking, and keep your ears in charge.
Updated · 7 min read · Lesson 6 of 14 in Audio Science
In short
- FFT bins sit sample rate divided by FFT size apart, and the block lasts FFT size divided by sample rate, so finer frequency detail always costs time.
- Windows trade leakage against peak width. Hann is a sensible default, and the low octaves always get few bins.
- Listen first, then set the analyzer for the question: large FFT for low notes, small for transients, and never EQ to match a reference curve.
In this lesson (8)
Matching the curve
You load a , put an analyzer on it and on your mix, and start EQing your master until the two curves look the same. Forty minutes later the shapes are close, and the mix sounds worse. The curve matches. The music does not.
The analyzer was answering a narrower question than the one you asked. To read it well, you need to know what its settings do to the picture.
Why it matters
The display is a set of choices
An analyzer takes a short block of samples, runs a fast () on it and draws how much energy each frequency had in that block. Then it takes the next block and does it again. The block length, the fade at its edges and the averaging between blocks all change what you see. The same mix can look smooth or jagged, fast or sluggish, and each view is accurate for its settings.
If the Fourier transform itself is new to you, start there. This lesson is about the analyzer built on it.
In the filter demo below, sweep the cutoff and watch how quickly the live spectrum follows.
Listen
Shape a sound with a filter
Chords and soft noise, or a real mix, through one filter, with the spectrum drawn live. Sweep the cutoff and raise the resonance.
Start with your speakers or headphones low.
The science
Bins, windows and averaging
The FFT, a fast way to compute the discrete Fourier transform (Cooley and Tukey, 1965), turns a block of samples into frequency slots called bins, spaced evenly:
The block itself lasts
so the two are tied together: . To see finer frequency detail you need a longer block, and a longer block blurs events in time. At 48 kHz, a 1024-point FFT has bins 46.9 Hz apart and covers 21 ms. An 8192-point FFT has 5.9 Hz bins and covers 171 ms. This is a limit of analysis itself, not a weakness of a particular plugin.
The bins are evenly spaced in hertz, but the display is logarithmic, with each octave the same width, so the low octaves get very few bins. With 4096 points at 48 kHz, the octave from 20 to 40 Hz holds about two bins, while the octave from 10 to 20 kHz holds more than 800. That is why the sub region looks blocky while the top end looks detailed. Two bass notes a semitone apart, E1 at 41.2 Hz and F1 at 43.7 Hz, are only 2.5 Hz apart. Telling them apart takes bins around a hertz wide, and that means a block close to a second long.
Windowing deals with the edges of the block. The FFT treats each block as if it repeated forever. A real block almost never starts and ends at the same point of the wave, so the repeat has a jump in it, and that jump spreads energy into bins where there is none. This is . A window fades the block in and out so the ends meet at zero. A rectangular window, which means no fade, gives the narrowest peaks and the most leakage. Hann, the usual default, leaks far less at the cost of wider peaks. Blackman-Harris leaks less still and widens peaks further (Harris, 1978).
Finally, averaging. A raw FFT display jumps with every block. Analyzers overlap their blocks and average them or slow the decay of the display, so a steady tone reads as a steady line. Heavy averaging shows the long-term balance of a mix. Light averaging with a peak hold shows what the are doing. Many analyzers can also tilt the display, often by 3 or 4.5 dB per octave, so that a typical mix reads roughly flat. Know which tilt yours uses before you judge the top end.
Try it
Resolve two tones
You need a with an FFT size setting. If your stock analyzer leaves it out, use a third-party analyzer that has one.
0 of 7 steps done
Nothing changes in the sound. Every difference you saw came from the analyzer settings.
Common mistake
Matching a reference curve
You a mix until its spectrum matches a reference, and no other mistake wastes as much time. The average spectrum of a finished song reflects its arrangement, its key and its instruments as much as its mix. Two great songs in the same genre have different curves because they contain different notes. Force yours onto someone else's and you boost gaps your arrangement left on purpose and cut energy it needs.
Use references for broad checks. "My mix has far more energy below 50 Hz than anything else in this genre" is worth knowing. "My mix is 1.3 dB lighter at 2.4 kHz than the reference" is noise. Spectrograms, the scrolling waterfall displays, have the same trade-offs painted in colour. They are good for spotting a hum that never moves or a that rings after each hit.
Takeaway
Listen, measure, decide
Use the analyzer to check an idea, not to find one. Listen first and name what bothers you: the vocal is buried, the low end booms. Guess where it lives, then open the analyzer and look there, with the FFT size set for the job, large for low notes and steady tones, small for transients. Make the change, match the level so the quieter version is not judged as worse, and decide by ear. If you are watching the analyzer more than you are listening, close it.
Reference track
A finished record you compare your mix against to check balance, tone and loudness on the same speakers.
Fourier transform
Maths that describes any sound as a sum of sine waves, each with its own frequency, level and phase.
FFT
Fast Fourier Transform. The algorithm analyzers use to split a short block of audio into frequencies. Longer blocks show frequency more precisely and timing less precisely.
Window function
A taper applied to each block of audio before an FFT, so the block edges do not smear energy into neighbouring frequencies. That smear is called spectral leakage.
Transient
The short, sharp start of a sound, like the crack of a snare or a pick hitting a string. It carries a lot of how hard a sound feels and how quickly the ear recognizes it.
Spectrum analyzer
A meter that shows how much energy sits at each frequency, usually on a log scale from 20 Hz to 20 kHz.
Resonance
A frequency where a system rings or builds up energy. On a filter, the resonance control boosts a narrow band around the cutoff.
3 questions
Check yourself
Sources
Show all 3Hide sources
- Cooley, J. W., & Tukey, J. W. (1965). An algorithm for the machine calculation of complex Fourier series. Mathematics of Computation, 19(90), 297-301.
- Harris, F. J. (1978). On the use of windows for harmonic analysis with the discrete Fourier transform. Proceedings of the IEEE, 66(1), 51-83.
- Smith, J. O. Spectral Audio Signal Processing. CCRMA, Stanford University. ccrma.stanford.edu/~jos/sasp/
Next lesson · 7 of 14
Filters are shape machinesThe slope of a filter decides how fast it cuts, and a normal EQ filter also shifts timing near its cutoff. Here is when that matters and when it does not.Previous: Fourier shows the ingredients

Written by Virzy Guns
Songwriter and producer, founder of Virzy Guns Production. Now building HealingWave.