Ultrasonic Leak Detector

Point your device at a pressurized air, gas or vacuum fitting and watch the live spectrum of the reachable 15–22 kHz leak tail (default 17–22 kHz). A leak radiates broadband turbulent hiss; this screener tracks the sustained high-frequency energy above your noise floor, shows a relative leak-confidence indicator, labels dominant peaks in Hz, and lets you isolate the leak band, run an audible scan, and export your detections as CSV.

This is a screening aid, NOT a real ultrasonic leak detector. Genuine air/gas leaks radiate broadband ultrasound concentrated around ~35–45 kHz, and industrial detectors use a dedicated 40 kHz heterodyne sensor to capture it. A normal phone/laptop microphone runs at 44.1 or 48 kHz, so by the Nyquist limit it cannot hear anything above ~22–24 kHz — it only catches the faint low-frequency tail of a leak that spills into the 15–22 kHz band. The confidence indicator and dB meter are relative heuristics (uncalibrated dBFS), not a calibrated measurement. Use this to flag obvious hissing leaks at the audible edge; for real surveys use a 40 kHz ultrasonic detector. Auto-gain, noise suppression and echo cancellation are requested off. Nothing is recorded or uploaded.

Idle — press Start to allow your microphone and screen for the audible edge of a leak.

Higher start frequency rejects more background; the leak tail is faint, so widen if you see nothing.
Records timestamped detections when band energy holds above your noise floor. The level meter and confidence are always computed inside the leak band.
Plays a pitch through your speakers that rises with the detected leak-band level. Mind your volume.
Relative leak confidence heuristic
Set your noise floor in clean air, then aim at the suspected leak.

Leak-band level
rel. dBFS (uncalibrated)
Above noise floor
dB over baseline
Dominant peak
strongest bin in band

Vertical axis = relative level (dBFS, uncalibrated). Horizontal axis = your selected leak band within the reachable 15–22 kHz range (the band labels under the axis show the live span). Amber dashed line = your set noise floor; the amber line at the right marks your mic’s Nyquist limit if it falls inside the view.

Detection log

Each time sustained leak-band energy holds clearly above your noise floor, the tool logs a timestamped detection with the dominant peak and the dB excess. Use it to compare suspected spots; export the list as CSV below. This is a relative ranking, not a calibrated leak rate.

TimePeak (Hz)Band dBFS+dB
Start the detector to log leak-band detections.

Microphone reach

A leak’s strongest energy sits near 40 kHz, far above what your mic can sample. Here is exactly how high your hardware can reach — everything above the Nyquist limit is invisible to this tool, no matter how loud the leak.

Sample rate
Nyquist limit (max audible to mic)
Frequency resolution
Leak-band coverage

Most built-in mics also roll off above ~16–18 kHz, so even the reachable tail is attenuated. If the spectrum is flat and silent in the band, the leak energy is simply out of reach — that is a hardware limit, not the absence of a leak.

How to Use It

  1. Quiet the area and press Start. Grant microphone access when prompted. Background machinery, fans and HVAC add broadband noise that masks a faint leak tail — the quieter the room, the better your odds at the audible edge.
  2. Set your noise floor. Aim the mic at clean air, away from the suspected leak, and press Set noise floor. The tool captures the current leak-band level as your baseline so the confidence indicator measures the excess energy a leak adds, not the absolute room level.
  3. Sweep slowly toward the fitting. Move the mic across joints, valves, couplings and seals. A leak is loudest a few centimetres from the orifice and falls off fast, so go slow and watch the leak-band level and confidence climb as you near it.
  4. Log detections and use the audible scan. Keep Log sustained leak-band detections on to record timestamped detections when band energy holds above your noise floor. Turn on Audible scan feedback to hear a pitch that rises with the level — useful for pinpointing by ear while you watch the screen. Watch your volume.
  5. Confirm, log and export. A real leak gives a sustained, repeatable rise that tracks the fitting — not a one-off blip. Detections are logged with a timestamp; press CSV to download the list. Always confirm with soap-bubble spray or a proper 40 kHz detector before acting.

Understanding Your Results

Leak-band level and “above noise floor”

The leak-band level is the average energy your mic captures inside the selected 15–22 kHz band, in relative dBFS (decibels relative to digital full scale). It is not calibrated sound pressure level — you cannot read a dB SPL or a leak rate from it. What is meaningful is the change: the “above noise floor” figure subtracts the baseline you set, so a steady positive excess that grows as you approach a fitting is the real signal of interest.

The leak-confidence indicator

Confidence is a relative heuristic derived from how far the sustained broadband band-energy sits above your noise floor, not a probability and not a calibrated verdict. Industrial practice evaluates a leak by its broadband level rather than a single tone, which is why this tool sums energy across the band instead of trusting one peak. Treat Possible/Likely as “worth a closer look,” and always confirm physically.

Dominant peak in Hz

The dominant-peak label marks the strongest frequency bin inside the band, refined to sub-bin accuracy. A true turbulent leak is broadband, so the peak wanders rather than locking to a fixed tone — a rock-steady single frequency is more likely electronic whine (a switching supply or coil) than a leak. The peak frequency itself is reliable; the level is not calibrated.

Microphone reach

The reach panel shows your sample rate, the resulting Nyquist limit (half the sample rate — the absolute highest frequency the mic can represent), and what fraction of the leak band that limit covers. Because a leak’s main energy lives near 40 kHz, well past any 22–24 kHz Nyquist ceiling, a negative result here never rules out a leak.

How It Works

When pressurized air or gas escapes through a small orifice, the flow becomes turbulent, and turbulence radiates a broadband acoustic hiss with strong energy in the ultrasonic range. Published industrial guidance puts a compressed-air leak’s strongest content around 35–45 kHz, with a stochastic spread that can reach roughly 20–100 kHz; the recommended way to evaluate a leak is by its broadband level, not a single frequency. Refrigerant and vacuum leaks behave the same way — commercial detectors sense the escaping-gas sound in a band near 38–42 kHz.

A purpose-built ultrasonic leak detector uses a 40 kHz heterodyne sensor: it captures that 40 kHz turbulence and mixes (“heterodynes”) it down into the audible range so you can hear the leak distinctly even on a noisy plant floor. Your phone or laptop has none of this. It samples audio at 44,100 or 48,000 Hz, and the Nyquist–Shannon theorem says it can only represent frequencies up to half the sample rate — about 22–24 kHz. The 40 kHz heart of a leak is simply not present in the data your browser receives.

So this tool deliberately works only in the 15–22 kHz band: the faint low-frequency tail of a leak that spills below the Nyquist ceiling. It runs a Fast Fourier Transform on each short window of mic audio, sums the relative energy inside the band you select, compares it against a baseline you capture in clean air, and converts that excess into a relative confidence reading and an audible scan pitch. The dominant bin is labelled in Hz; sustained detections are logged with timestamps for CSV export.

Published leak signatures (typical reference values)

These are widely-cited typical figures from leak-detection literature, not guarantees — every leak’s spectrum depends on orifice shape, pressure and gas. The point of the table is to make clear how far the real energy sits above your mic’s reach.

Where leak energy actually lives, versus what a browser microphone can capture.
SourceTypical ultrasonic signature
Compressed air / nitrogen leakBroadband turbulent hiss, strongest ~35–45 kHz, spread ~20–100 kHz; evaluated by broadband level.
Refrigerant / pressure / vacuum leakSame turbulent mechanism; commercial detectors sense it near 38–42 kHz.
Industrial ultrasonic detectorDedicated 40 kHz (±2 kHz) heterodyne sensor; converts ultrasound down to audible.
What this browser tool can seeOnly the 15–22 kHz tail below the mic Nyquist (~22–24 kHz). The 40 kHz core is unreachable.

What is trustworthy and what is not

The frequency of a detected peak is reliable — it comes straight from the FFT. The presence of sustained broadband energy in the band, rising as you approach a fitting, is a genuine clue. What is not trustworthy is anything implying calibration: the level is relative dBFS, not dB SPL, the confidence is a heuristic, not a probability, and there is no leak rate. A silent band never rules out a leak, because the leak’s real energy is above the Nyquist limit. For surveys that matter — energy audits, safety, refrigerant compliance — use a 40 kHz ultrasonic detector and confirm with soap-bubble testing.

Frequently Asked Questions

Can my phone or laptop really detect an air leak?
Only partly, and only the easy ones. A real leak radiates broadband turbulence strongest around 35–45 kHz, but a browser microphone samples at 44.1 or 48 kHz, so by the Nyquist theorem it cannot represent anything above about 22–24 kHz. This tool works only in the 15–22 kHz tail that spills below that ceiling, so it can flag a loud, close hissing leak but will miss the 40 kHz core that an industrial detector hears. It is a screening aid, not a substitute for a 40 kHz ultrasonic leak detector.
Why does it only look at 15–22 kHz instead of 40 kHz?
Because 40 kHz is physically out of reach for your hardware. The Nyquist–Shannon sampling theorem limits any digital system to frequencies below half its sample rate, so a 44.1 kHz mic tops out near 22 kHz and a 48 kHz mic near 24 kHz. The 15–22 kHz band is the highest slice of a leak's broadband tail that your mic can actually sample. The Microphone reach panel shows your exact Nyquist limit.
Is the leak-confidence indicator a real measurement?
No. It is a relative heuristic, not a calibrated measurement or a probability. It looks at how far the sustained broadband energy in the leak band sits above the noise floor you set, and maps that excess to a Possible / Likely style verdict. The underlying level is uncalibrated dBFS, not dB SPL, and there is no leak rate. Always confirm a suspected leak with soap-bubble spray or a proper instrument.
Why do I need to set a noise floor first?
A leak shows up as extra energy on top of whatever your room already produces. By capturing the leak-band level in clean air first, the tool measures the excess a leak adds rather than the absolute level, which makes the confidence reading meaningful in your specific environment. Re-set the noise floor whenever you move to a noisier or quieter spot.
What does the audible scan feedback do?
It plays a tone through your speakers or headphones whose pitch rises as the detected leak-band level increases, so you can hunt for the loudest spot by ear while watching the screen. It is loosely inspired by how industrial detectors heterodyne ultrasound down to audible pitch, but here it is just a level-driven tone, not a true frequency translation. Start at low volume and raise it carefully.
Why must auto-gain and noise suppression be off?
The tool requests the raw microphone signal with automatic gain control, noise suppression and echo cancellation switched off, because those features reshape the spectrum — pumping levels and carving out frequencies — which would corrupt a faint leak tail. We request them off, but a browser or OS may still apply processing it does not expose, so if the band looks artificially clean, disable microphone enhancements in your system settings.
Is my audio recorded or uploaded?
No. The microphone signal is analysed in real time entirely in your browser to compute the spectrum and confidence, then discarded — nothing is recorded, saved, or transmitted. The CSV export is a local download you trigger yourself. The microphone is released the moment you press Stop or close the tab.