Leak noise correlator on your phone: locating a leak between two sensors by arrival time
You have a 25-metre buried line under a tiled courtyard, and the meter has proved it is leaking. The ground microphone gives you a wide area, and the customer does not want the whole courtyard broken up.
This guide explains the leak noise correlator in Kashf Pro. It is an experimental tool that takes sound from two sensors at two points on the same pipe, calculates the difference in the time the leak noise takes to reach each one, and estimates the leak position in metres from sensor A. We explain honestly what equipment you need, when to trust the number and when to reject it.
What is a leak noise correlator and what does it give you?
A correlator compares two signals recorded at the same moment at the two ends of a pipe section. Leak noise reaches the nearer sensor first, and the time difference between the two arrivals gives the position.
The Two-Sensor Correlator in the Kashf Pro app records 10, 20 or 40 seconds from two channels through a single audio interface, then shows:
- The estimated position in metres from sensor A.
- The arrival time difference in milliseconds.
- A confidence percentage.
- A drawing of the pipe section between A and B, with a red dot at the position.
You can save the result in a report.
Who is this tool for?
For technicians who own, or can put together, two contact or ground sensors and a two-channel audio interface, and who know the pipe's route, length and material.
It is also useful for companies that want to try acoustic correlation before buying a dedicated correlator. It is not a beginner's tool, and it does not work with a single phone and its built-in microphone.
When do you use it?
- After the leak has been proved with a meter test or a pressure test, and the suspect section has been identified.
- On a long buried line between two exposed access points, such as the meter and the tank valve, two valves, or two taps.
- When listening from the surface is weak because of depth or the type of floor.
Where do you use it?
On pressurised supply pipes with two reachable contact points. It suits the line from the meter to a ground tank, courtyard and garden lines, and lines between a ground tank and a roof tank, a common arrangement in Saudi homes.
The result is clearer on metal pipes, because they carry sound further. Plastic absorbs sound quickly, so keep the sensors closer together.
What do you need before you start?
- A two-channel (stereo) USB audio interface that works with your phone through USB-C or a suitable adapter. Both channels must be recorded together by one device so the timing is synchronised. Two separate phones will not work.
- Two separate sensors, such as contact microphones, ground microphones or vibration sensors, with an amplifier if the sensor needs one. A goes on the left channel and B on the right.
- Long enough cables to connect both sensors to the interface across the whole distance.
- The actual pipe length between the sensors along its route, not the straight-line distance, and its material.
- A pressurised line, the pump and air conditioners switched off, and a Kashf Pro account whose plan includes this tool or that holds credits.
How to use it correctly, step by step
Two-Sensor Leak Correlator (experimental)Subscription — free trial available
The tool records both channels together at the device's sample rate, usually 48 kHz. It then applies the GCC-PHAT algorithm, generalised cross-correlation with phase transform.
It splits the recording into segments of about two-thirds of a second, with 50% overlap and a Hann window. It calculates each channel's spectrum and keeps only frequencies between 100 and 2500 Hz. It then divides the cross-spectrum by its magnitude so that only the phase remains, and transforms it back into time. The highest peak within the physically possible delay, meaning length ÷ speed, gives the arrival time difference Δt.
The formula:
Distance from A = (length + speed × Δt) ÷ 2
A positive Δt means the sound reached A later, so the leak is closer to B.
Confidence = 1 − (second-highest peak ÷ highest peak). A result is accepted when confidence is above 30%.
Before calculating, and before any session is counted against your plan, the tool checks for three situations and rejects them with a clear message rather than a misleading number:
- The input has only one channel.
- The two channels are almost identical, meaning one microphone duplicated.
- One of the channels is silent.
The preset speeds are approximate and editable: steel / cast iron 1200 m/s, copper 1150, PPR/PVC about 450, and polyethylene (PE) about 350.
Steps for correct use
- Fit both sensors to the same pipe
Fix sensor A to an exposed metal contact point at the start of the section, such as a valve body, the meter or a tap. Fix sensor B to a point at the other end. Contact must be direct and firm, using a magnet or steady pressure, not through flaking paint or soil. Connect A to the left channel and B to the right.
- Choose the input and enter the length
Open the tool and choose the interface under Audio input (stereo). In Distance between sensors (m) enter the actual pipe length along its route; the default is 20 m. If the line bends around the corner of the courtyard, measure the route, not the straight line between the two points.
- Set the material and the speed
Choose the Pipe material, and Speed of sound (m/s) fills in with the default value. If you know a more accurate speed for your pipe from the manufacturer's specification or an earlier calibration, type it in. The speed in plastic varies with diameter and wall thickness, and any error in it carries straight into the position.
- Record in quiet conditions
Choose the Duration: 20 seconds by default, and 40 seconds for a faint leak or a noisy site. Make sure the line is pressurised and the pump is off if possible, then tap ⏺ Record and calculate. A session is counted only after the recording finishes and passes the channel check, so a mono, duplicated or silent input costs nothing. If you leave the screen while recording, the microphone stops immediately. Do not touch the cables or walk near the sensors while recording.
- Read the result and repeat
If a position appears with confidence above 30%, record once or twice more. Results that agree are a good sign. Then move one sensor to shorten the section around the point and measure again. Save the result with 💾 Save result to get a report link.
How to read the result
“Estimated position: X m from sensor A” in green: the correlation is clear and confidence is above 30%. Below it you see the arrival time difference in milliseconds and the confidence percentage, and the red dot on the section drawing.
- “Source probably outside the distance between the sensors (on sensor A's or B's side)”: shown when the arrival difference equals the travel time over the whole distance (|Δt| × speed ≥ length), so no position inside the section is displayed. Move both sensors towards the side named so the suspect area falls in the middle.
- Position at a tee: the leak may be on the branch itself, because its sound enters the main line at the junction.
- “Weak correlation (…%) — no clear shared leak sound, or the noise is high”: do not rely on the number. Increase the duration, shorten the distance, improve the contact or switch off noise sources. Also make sure the line is pressurised.
- “The selected input has only one channel”: the interface is not selected or does not expose two channels to the browser.
- “The two channels are almost identical”: you connected one microphone to both channels.
- “One of the channels has no signal”: a sensor or cable is disconnected.
The save button only works with confidence above 30%.
The physics behind the number: why speed and length matter most
If the leak is at distance d from A and the total length is L, the sound travels d to reach A and L − d to reach B. The time difference between them is Δt = (2d − L) ÷ v, so d = (L + v × Δt) ÷ 2.
Timing resolution is not the problem. One sample at 48 kHz equals about 0.02 milliseconds, which in a PPR pipe corresponds to only a few millimetres. The real error comes from the speed and the length.
Example
A 30 m PPR pipe, a real leak 10 m from A, and an actual sound speed of 400 m/s.
- The real arrival difference = (20 − 30) ÷ 400 = −25 milliseconds.
- If you leave the default speed of 450 m/s, the tool calculates (30 + 450 × −0.025) ÷ 2 ≈ 9.4 m.
- The error is about 60 cm, caused by the speed alone.
An error of one metre in the length shifts the result by half a metre. So measure the route with a tape; do not judge it by eye.
To calibrate the speed in the field, open a tap at a known position between the sensors slightly to simulate a leak, then adjust the speed until the result matches its real position. Then close it and measure the leak.
Correlator or listener? When to choose each
- The listener works from the surface, compares points on a grid and needs only one device. It works well when the line is close to the surface and the floor carries sound.
- The correlator does not need you to stand over the leak, but it needs two contact points on the same pipe, a known length and material, and a two-channel interface. It suits long buried lines under courtyards and roads.
- Good practice: use the correlator to narrow the section down to a few metres, then confirm from the surface with the listener or tracer gas before digging.
Common mistakes that spoil the result
- Straight-line distance instead of route length: the most common mistake. Pipes rarely run in a straight line.
- Mixed materials in the section: part old steel and part new PPR means two speeds, and a single speed gives a skewed position. Split the measurement where the material changes.
- Swapped channels: if A is connected to the right channel the position is mirrored, so you get L − d instead of d.
- A shared noise source: a pump or a whistling valve on the same line reaches both sensors and produces a correlation peak in its own direction, not the leak's.
- A sensor on another pipe: contact with a neighbouring pipe or with reinforcing steel instead of the intended line.
- Poor contact or a moving cable: adds uncorrelated noise and lowers confidence.
The honest limits of the tool
- The tool is experimental and has not been field-tested at scale. Testing the algorithm on simulated signals estimated the delay accurately, but a real site has noise, attenuation and branches.
- It needs two separate sensors and a two-channel audio interface. It does not work with one phone, or with two separate phones, because their timing is not synchronised.
- The default speeds are approximate. The real speed changes with diameter, wall thickness, pressure and temperature.
- The band used is 100 to 2500 Hz. If the leak noise is very faint, or the distance is long on plastic, no correlation may appear.
- A source outside the section: the tool detects this when the arrival difference reaches the full travel time and says so plainly, but a source very close to one sensor can give the same result.
- Some plans do not include this tool, and it can be used with credits.
- The position is an estimate that supports your judgement. Do not dig on a single measurement; repeat it and confirm from the surface.
When to escalate or ask for a second opinion
- Confidence stays weak despite better contact and a longer duration, or repeated results are far apart.
- A long plastic line under asphalt or expensive tiling. Tracer gas from the surface helps confirm the position before digging.
- The leak is near foundations, a boundary wall or a ground tank. A licensed structural engineer may be needed to assess the effect on soil and concrete.
- The customer needs a technical report from a detection company.
You can ask the Kashf team, which works in Saudi Arabia, for a second opinion or subcontract them. The team detects leaks electronically without breaking, using acoustic and thermal equipment, pressure testing and tracer gas, and hands over a written report that can be checked on the verification page. Book from the order page or on WhatsApp at +966544027180.
Frequently asked questions
Can I use the leak noise correlator with two phones instead of an audio interface?
No. The tool calculates a time difference in fractions of a millisecond, which requires both channels to be recorded on one clock in one device. Two separate phones cannot be synchronised that precisely. That is why the tool requires a two-channel USB audio interface with both sensors connected to it.
What is the speed of sound in PPR and steel pipes?
The default values in the tool are approximate: about 1200 m/s for steel and cast iron, 1150 for copper, about 450 for PPR and PVC, and about 350 for polyethylene. The real speed varies with diameter and wall thickness, so it is better to calibrate it with a tap at a known position when you can.
Why does the tool say the source is outside the distance between the sensors?
When the arrival difference reaches the travel time over the whole distance, the tool shows “Source probably outside the distance between the sensors” instead of a number. It means the sound source is at one of the sensors or outside the section, or that noise such as a pump enters from one end. Reposition the sensors so the suspect area is in the middle and switch off noise sources.
What does weak correlation mean in the result?
It means the highest peak in the correlation function does not stand out much from the others, so there is no clear shared sound between the two sensors. The cause is usually a faint leak, a long distance on plastic, poor contact, high noise or a line without pressure. Do not rely on the number in that case.
Does the tool replace a professional correlator?
No. It is experimental and uses general-purpose audio hardware. Dedicated units have calibrated sensors, synchronised wireless transmission and advanced filters. The tool helps you experiment, learn and handle clear cases, and the final decision rests with your experience and confirmation from the surface.
How far apart should the two sensors be?
There is no fixed figure; it depends on the material and the strength of the leak. Metal allows longer distances than plastic. Start with a short section between two access points; if confidence is weak, shorten the distance, and if a clear position appears, narrow it down with a second measurement.
Does it work on drain lines?
Not well. The leak noise the tool relies on comes from pressurised water escaping through a crack, and drain lines are usually not pressurised. Drain lines are checked with other methods, such as an internal camera or a fill test.
Is the tool available on every plan?
No. Each plan has per-tool limits set by the platform administrator, and this tool may be unavailable on one plan and unlimited on another. Each plan's limits are shown on the plans page, and you can use usage credits if your plan does not include it.
This guide is educational. It explains estimation tools that help you understand a situation and decide what to do next; confirming a leak and pinpointing it always requires an on-site inspection with instruments. Structural questions are for a licensed engineer to decide.