🔍

Noise Type Identifier

Capture ambient noise with your microphone, run an FFT, and classify the dominant type from its spectral shape — hum, buzz, hiss, rumble, rattle, broadband or pure tonal — then see a likely source (electrical, mechanical, airflow or structural) with practical remediation tips.

What’s reliable vs. what’s a guess. The type is read from the spectral shape (slope, tonal peaks, 50/60 Hz content) — that is calibration-independent and trustworthy with the same mic and setup. The source suggestion is a heuristic starting point, not a diagnosis. Any level shown is an uncalibrated estimate from a consumer mic — not a certified dB SPL and not legal/complaint evidence. Keep automatic gain control and noise suppression OFF (this tool requests a raw stream) or the reading is meaningless. A consumer mic also rolls off at the extremes and generally cannot capture true infrasound (<20 Hz). Audio is analysed live and never recorded or uploaded.

Microphone

Idle — press Start to begin.

Verdict

Start listening to classify the dominant noise type.
Likely source (heuristic)
A starting point for troubleshooting — confirm by switching suspected sources off and watching the spectrum change.

Live spectrum (log frequency)

Spectral slope
Dominant peak
Mains content
Level

Optional: calibrate the level (shared)

Calibration is optional and only affects the rough SPL estimate (marked *). Classification works without it. The offset is stored locally and shared with every Noise Analysis tool, so you only calibrate once.

How It Works

When you press Start, the tool opens your microphone requesting a raw stream with automatic gain control, noise suppression and echo cancellation off (the browser or OS can still override this) and feeds it into a Web Audio FFT analyser. Every frame it measures the shape of the spectrum rather than just how loud it is — and shape is what tells different noises apart.

Three calibration-independent features do most of the work. The spectral slope (in decibels per decade of frequency) says whether energy falls toward the highs (rumble, pink-ish), rises toward the highs (hiss), or stays flat (broadband/white). A tonality measure and the prominence of the strongest peak above the spectral median tell whether the sound is a narrow tone (hum, pure tonal) or harmonic-rich (buzz). The crest factor of the waveform — peak versus RMS — flags intermittent, spiky rattle against a steady drone. Finally the tool measures how much energy sits within a couple of hertz of 50/60 Hz and their harmonics (100/120, 150/180 Hz), the fingerprint of mains-related hum.

From those features it picks the best-matching type and shows a rough shape-match percentage. It then maps the type to a likely source category: a 50/60 Hz family points to electrical (transformer, LED driver, ballast, ground loop); a tonal mid-frequency sound with harmonics to mechanical (fan, motor, bearing); rising high-frequency hiss to airflow (HVAC, vents); and broadband low-frequency rumble to structural (traffic, building plant). That source line is a heuristic, not a diagnosis — the honest way to confirm it is to switch a suspected source off and watch the spectrum change, since a before/after comparison with the same mic is genuinely meaningful even when the absolute level isn’t calibrated.

The optional calibration turns the relative dBFS level into a rough dB SPL estimate. Read the current level on a real sound-level meter (or a calibrated phone app), type that number in, and the tool stores the difference as an offset. That offset is shared across every Noise Analysis tool via your browser’s local storage, so you calibrate once. Even calibrated, the estimate is approximate and is not a substitute for a Type 1/2 meter or valid as legal evidence.

Frequently Asked Questions

How accurate is the noise-type classification?
The type is read from the spectral shape — its slope, tonal peaks, crest factor and 50/60 Hz content — which is calibration-independent and reliable as long as automatic gain control and noise suppression are off (this tool requests a raw stream). It works best on a steady sound you can hold for a few seconds; rapidly changing or mixed noises are harder, and very quiet sounds near the mic’s noise floor will read as “too quiet.”
Is the “likely source” a diagnosis?
No. It is a heuristic starting point that maps the noise type to a common source category — electrical, mechanical, airflow or structural — with general fix-it tips. Many real situations are mixtures, and similar shapes can come from different sources. Use it to decide what to check first, then confirm by switching a suspected source off and watching the spectrum change.
Why must I keep automatic gain control and noise suppression off?
Those features continuously alter the signal — boosting quiet passages and carving out steady tones — which is exactly the spectral information this tool measures. With them on, a hum could be suppressed and a slope flattened, making the reading meaningless. The tool asks the browser for a raw stream with AGC, noise suppression and echo cancellation disabled; if your OS or browser overrides that, results will be unreliable.
Can it measure infrasound or very deep rumble?
Not reliably. Consumer microphones roll off sharply at the frequency extremes and generally cannot capture true infrasound below about 20 Hz, so the tool ignores content below 20 Hz. It can still classify audible low-frequency rumble, but treat anything at the very bottom of the range with caution and don’t expect a sub-20 Hz reading.
What does the calibration do, and is the dB SPL number trustworthy?
A browser mic is uncalibrated, so the tool shows relative dBFS by default. Calibration lets you read a real meter once, enter that level, and store the difference as an offset that every noise tool reuses. The resulting dB SPL figure (marked with a *) is an approximate estimate only — it is not a certified measurement, not a substitute for a Type 1/2 sound-level meter, and not valid as legal or complaint evidence. Classification does not need calibration at all.
Is my audio recorded or uploaded?
No. The microphone signal is analysed in real time in your browser to compute the spectrum and is never recorded, saved, or transmitted. The microphone is released when you press Stop or close the tab, and the only thing stored locally is your optional calibration offset.