Before reading anything off a spectrum, note the window, the averaging and the vertical scale, because each of the three can change the curve's shape enough to invert your conclusion.
Read a spectrum analyser without being lied to
checked 2026-09-16responsible Room Tone Register (a named register, not a person)
The three settings that change the answer
| Setting | What it does | Common mistake |
|---|---|---|
| FFT size | Sets frequency resolution and time window | Too small hides closely spaced tones |
| Averaging | Smooths the trace over time | Too much averaging hides a cycling source |
| Vertical scale | Log or linear | A linear scale makes a 10 dB peak look enormous |
The three ways it lies
- It shows level, not audibility. A 40 dB peak at 8 kHz is inaudible under a voice; a 25 dB peak at 120 Hz is not. Weight the curve mentally against the ear before you react to it.
- It shows the microphone as well as the room. A measurement microphone's own response and its housing diffraction are in the curve. Compare against the microphone's own calibration file before blaming the room.
- It shows the moment, not the room. Without enough averaging you are reading one instant of a cycling source; with too much you are reading an average that never occurs.
The order to work in
Set the resolution from the question you are asking, not from the default. If you are hunting a tone, use a large FFT and look for narrow peaks. If you are judging the balance of a room, use octave or third-octave smoothing so the fine structure does not distract you. Then average long enough to cover at least one full cycle of the slowest source you suspect — which for a domestic compressor means several minutes, not several seconds.
A waterfall or spectrogram plot answers the question a single spectrum cannot: what changes over time. If the complaint is 'it comes and goes', the spectrum is the wrong display and you should switch before you spend an hour on it.
Source log
- 01Fourier Analysis of a Waveform, Penn State Acousticssupports the Fourier relationship between window length and frequency resolution
- 02Energy Decay Relief, CCRMA (Stanford)supports the time-frequency display used when a single spectrum is insufficient
- 03Phase Diagrams, Penn State Acousticssupports the phase information a magnitude spectrum does not show
Each line points at one specific document. No line is a home page.
Leave this page Close this page once you have the resolution and averaging noted. Without those two numbers the curve is not evidence.