This tool breaks your noise floor into octave bands — the standard way acousticians describe the shape of noise. Each band is one octave wide (its upper edge is double its lower edge), forming the ten standard 1/1 octave centres at 31.5, 63, 125, 250, 500 Hz, 1, 2, 4, 8 and 16 kHz. For every band the analyzer sums the power in that frequency range from an FFT and reports a level, so you can see whether your noise is dominated by low-frequency rumble, midrange, or high-frequency hiss.
Instead of a jumpy instantaneous reading, it computes Leq — the equivalent continuous level, the average power since you started (or last reset). That’s the meaningful way to characterise steady noise, and the same quantity used in standards like ISO 1996 for environmental noise assessment. The per-band powers are calibrated so they sum to the broadband total, which is measured directly from the waveform in dBFS. The broadband Leq and A-weighted Leq figures give you two complementary views: total physical energy, and perceived energy weighted toward how the human ear responds around 2–4 kHz. A large gap between the two usually means your noise is heavily low-frequency (HVAC rumble, traffic), which is common in rooms with poor low-frequency isolation. For tracking a microphone noise floor or diagnosing spectral noise from a preamp, the per-band breakdown reveals exactly which frequency region is the problem.
Flat (Z) vs A-weighting
Flat (Z) shows the true energy in each band. A-weighting applies the standard curve that approximates how sensitive human hearing is at each frequency — it heavily discounts low frequencies and slightly boosts the 2–4 kHz region. A-weighted noise is closer to how annoying the noise sounds: a floor dominated by sub-100 Hz rumble looks high on the flat view but drops a lot when A-weighted, because you don’t hear low frequencies as loudly.
Reading the result
- Dominant band — the band with the most energy (in the current weighting). Low-band dominance ⇒ rumble/HVAC/handling; high-band ⇒ hiss/fans; a single mid band ⇒ a tonal source like an electrical hum at 50 or 60 Hz.
- Broadband Leq — the overall averaged floor in dBFS. Lower is better; what matters is the gap to your signal — a classic measure of signal-to-noise ratio. If you need to calculate that gap numerically, pair this tool with the signal-to-noise ratio calculator.
- A-weighted Leq — the overall floor as perceived. If it’s much lower than the flat figure, your noise is mostly inaudible low-frequency energy. This is especially common in home recording studio noise floor measurements where HVAC dominates the low bands but sounds quiet.
How this differs from the Background Noise Detector
The Background Noise Detector is the quick consumer check — a single quiet/moderate/noisy verdict plus hum hunting. This analyzer is the technical view: octave-band Leq, A vs Z weighting, and the spectral distribution of your floor. Use that one to spot hum fast; use this one to characterise the noise precisely. For occupational contexts — converting a dBFS-derived level estimate into a noise dose against OSHA or NIOSH limits — the noise exposure calculator takes level-and-duration pairs and outputs a compliance picture. If you find a strong tonal component at a specific frequency in one of the octave bands and want to generate a test tone to confirm it, the frequency generator lets you produce a reference tone at any frequency in your browser.
What is Leq and why use it?
Leq is the equivalent continuous level — the constant level that carries the same energy as the fluctuating signal over the measurement period. For steady noise it’s far more stable and meaningful than an instantaneous reading, which is why this tool averages power over time rather than showing a single snapshot. Let it run for several seconds, then read the Leq.
Why dBFS and not dB SPL?
A browser can’t access calibrated sound-pressure levels, so the analyzer measures the digital signal level (dBFS), not acoustic loudness in dB SPL. The shape across bands and the A-vs-Z comparison are meaningful and calibration-independent — but the band shape still reflects your microphone’s own frequency response (most consumer/laptop mics roll off at the low and high extremes), so read it as indicative, not a lab-grade spectrum. The absolute numbers are relative to full scale and depend on your input gain, so don’t read them as room SPL.
What’s the difference between A-weighting and flat?
Flat (Z) shows the actual energy in each band. A-weighting applies a frequency curve matching human hearing sensitivity — discounting lows, slightly lifting 2–4 kHz. A-weighted levels better reflect how loud/annoying the noise is to a listener; flat levels better reveal the physical energy distribution (useful for chasing rumble).
Why are octave bands used instead of a fine spectrum?
Octave bands summarise the noise shape into ten meaningful slices that map to how we describe sound (lows, mids, highs). They’re standard in acoustics, average out the randomness of broadband noise, and make it easy to see which region dominates — without drowning you in thousands of FFT bins. For a fine spectrum, use the Audio Spectrum Analyzer.
Should I stay silent during the measurement?
Yes — the point is to characterise the noise floor, the level when nothing is happening. Any speech or sound you make is averaged into the Leq and will skew the result. Stay quiet; if you accidentally make noise, press Reset average and start again.
My low bands are very high — is that bad?
Not always. Low-frequency energy (rumble from HVAC, traffic, handling) is common and often inaudible, which is why it shrinks under A-weighting. If the A-weighted figure is acceptable it may not matter for listening — but it can still eat headroom and trip up compressors, so a high-pass filter is often worthwhile.
Does it disable noise suppression?
Yes — it requests the raw signal with noise suppression and auto-gain off so you measure the true floor. If your OS forces processing that can’t be disabled, the bands will look artificially clean; turn off microphone "enhancements" in system settings for an accurate analysis.
Is any audio recorded?
No. The signal is analyzed in real time to accumulate band-power statistics only; nothing is recorded, saved, or transmitted. The microphone is released when you press Stop or close the tab.
How long should I average to get a stable noise floor reading?
For broadband noise (HVAC, fans, traffic) 10–30 seconds is usually enough to get a stable Leq. If your noise floor has occasional transients — a distant vehicle, a creak — let it average for 60 seconds or more, then use the Reset button if a loud event skewed the result and start again. The “Avg time” cell shows elapsed averaging time so you can judge stability. For a home studio noise floor measurement meant to compare against NC or NR curves, 30 seconds of quiet averaging per measurement is a common minimum.
What is a good noise floor level for recording or podcasting?
This tool reports dBFS (relative to digital full scale), not absolute SPL, so direct comparisons to acoustic targets like NC-25 require a calibrated meter. As a practical guide: if your A-weighted Leq sits below roughly −50 dBFS and you record at a healthy level (−18 to −12 dBFS RMS), the noise is unlikely to be audible in a finished mix. If the floor is above −40 dBFS A-weighted you will likely hear it. Dominant mid or high bands (250 Hz–4 kHz) are more problematic than low-frequency rumble because A-weighting doesn’t discount them as much.
Can I use this tool to check a USB audio interface or external mic, not just a laptop mic?
Yes. Any audio input your operating system exposes as a microphone device will appear in the Microphone drop-down once you grant browser permission. Select your USB interface or external microphone from the list. Keep in mind that the result reflects the combined noise floor of the acoustic environment, the microphone capsule, the preamp, and the analog-to-digital converter — so it is a system noise floor, not the mic’s self-noise spec in isolation. To isolate preamp noise, use a calibrated loopback or close the mic capsule with a cap during measurement.