Understanding Frequency Spectrums in Audio Data Analysis

What a Frequency Spectrum Actually Shows You

A frequency spectrum takes a short slice of a recording — often a fraction of a second — and shows how much energy is present at each frequency within that slice. The horizontal axis runs from low to high, usually 20 Hz to 20 kHz for music, and the vertical axis shows level in decibels. Where you see a tall peak, that frequency is loud in the material. Where you see a valley, it is quiet.

Most spectrum displays are built from a mathematical routine called a Fast Fourier Transform, which converts a chunk of samples into frequency information. That chunk is called the window or block size. A larger block gives finer frequency detail but blurs timing; a smaller block does the opposite. Knowing this matters, because two analysers showing different pictures of the same recording are often just using different block sizes.

Reading the Display Without Guessing

Start by identifying the shape rather than chasing individual spikes. A broad hump between roughly 100 Hz and 300 Hz usually means boxiness or mud in a mix. A narrow, unwavering spike at 50 Hz is almost always mains hum, which is the frequency of UK mains electricity — and if you see a smaller spike at 100 Hz, 150 Hz and 200 Hz above it, you are looking at its harmonics.

A few terms are worth keeping straight:

• Peak — a single frequency sitting noticeably above its neighbours, often a resonance or a tonal note.
• Noise floor — the flat, low-level haze across the whole display, made up of hiss, room tone and equipment noise.
• Harmonic series — evenly spaced peaks above a fundamental, which is what most instruments and voices produce naturally.
• Broadband noise — energy spread evenly across many frequencies rather than concentrated in one place.

If a peak moves up and down with the music, it is part of the performance. If it stays put when the performer stops, it is something in the room or the signal chain.

The Frequencies That Cause Most Trouble

Experience gives you a mental map of where problems tend to live. It is not a set of rules, but it is a useful starting point when you are staring at a suspicious lump in the display.

• Below 40 Hz — traffic rumble, footsteps through floorboards, handling noise on a microphone stand, air conditioning.
• 50 Hz and its harmonics — earth loops and poorly screened cables. Try lifting the earth on a DI box or moving a laptop power supply away from a microphone cable.
• 100–300 Hz — proximity effect on close-miked vocals, room resonance, and the general thickening that makes a mix feel heavy.
• 2–5 kHz — the presence region, where intelligibility lives, but also where harshness and listener fatigue creep in.
• 6–10 kHz — sibilance on vocals and cymbals, plus the whine of some computer fans and switch-mode power supplies.

Spotting a peak in one of these bands is not automatically a problem. A bass guitar with strong energy at 80 Hz is doing its job. The question is always whether that energy is wanted.

A Practical Workflow for Spotting Noise

Begin with a section of the recording where the performer is silent but the microphones are still open. That is your noise floor, and it tells you what you are fighting. Freeze the display or take a mental snapshot, then play the full take and watch which parts of the spectrum rise above it.

Next, listen while you watch. Solo the suspect band with a narrow filter and sweep it slowly. When the unwanted sound jumps out, you have found it. Narrow peaks usually need narrow cuts; broad tonal imbalances are better handled with gentle shelving rather than surgical notching, which can leave an audible hole.

Finally, compare before and after. A cut of 3 dB or less is often enough. If you find yourself reaching for 10 dB, the problem is probably better solved at the source — moving the microphone, changing the room, or replacing a noisy cable.

Settings That Change What You See

Analyser settings are not neutral. Averaging smooths the display and makes trends easier to read, but it hides short transients. Peak hold keeps the highest level at each frequency, which is excellent for catching intermittent noises such as a chair creak or a mobile phone buzz. A linear frequency scale is easy to read at the low end; a logarithmic scale matches how we hear and is more useful for musical decisions.

It is also worth remembering that the spectrum tells you nothing about timing. Two sounds can share an identical spectrum and still be completely different — a snare hit and a burst of noise, for example.

When Your Ears Should Overrule the Graph

The display is a map, not the territory. It is brilliant at confirming what you have already noticed and at catching problems buried beneath a loud mix, but it cannot tell you whether a recording moves people. Use it to diagnose, then put it aside and listen at a sensible level on the speakers or headphones you know best. If the spectrum looks tidy but the result sounds thin, dull or harsh, trust the sound. The graph will still be there tomorrow.