IARI Labs — Acoustic intelligence for industry

Watch

Real footage. No simulations.

Everything below was recorded on our own prototype hardware.

Acoustic camera prototype locating and naming a sound in the lab

NEW · OCT 2026

Locate it, then name it

The 60‑microphone prototype finds a sound in the room and labels it live: vacuum, alarm, rain. Beamforming runs on the FPGA; a neural network on the processor's NPU recognises the sound signature.

Live camera image with the sound map overlaid
Source position tracked at 23 maps per second
Classification on the device, every half second
One cable out: H.264 video stream

60‑microphone beamforming

The full hexagonal array tracking a source in real time.

Where it started: 8 microphones

The first proof of concept of our beamforming pipeline.

How it works

A camera that sees sound

Three stages, all inside one device. The output is a picture of your floor with every sound source located and labelled.

Hexagonal 60-microphone array PCB designed by IARI Labs

Capture with 60 ears

A hexagonal array of 60 MEMS microphones, designed in‑house, listens to the whole scene at once, from a hum at a few hundred hertz to a hiss above 10 kHz.

Sound map overlaid on the camera image: a speaker lights up in yellow

Locate in hardware

An FPGA combines all channels into a sound‑power map 23 times a second and overlays it on the camera image. The bright spot is the source; the crosshair is its peak.

The prototype labelling a located sound as VACUUM

Classify at the edge

A compact neural network on the processor's NPU recognises the sound signature and tags it on screen. Nothing leaves the device unless you want it to.

60 MEMS mics→FPGA beamformer→Edge SoC + NPU→Sound map · class · position→Alerts & integrations

Interactive demo

Put a source on the floor. Watch it light up.

A simulated plant driven by the same ideas as our prototype. Five arrays on the walls, time‑synchronised, point at the same event. Click a machine or move your cursor over the scene, and watch the sound map, the spectrum and the classifier respond.

Simulation for illustration. The videos above show the real device. Open the demo full screen ↗

Illustration of a plant floor with acoustic arrays on the walls, each covering a zone, and an acoustic alert on a robot joint

Plant scale

One array sees a cell. A synchronised network sees the plant.

Arrays mount on walls and columns, each covering a zone. Because they share one clock, every event is heard by several arrays at once and located where their beams cross.

Zone coverage that grows one array at a time, with no wiring into the machines.
Triangulation across arrays for a position on the floor, not just a direction.
Alerts with a picture: which machine, which joint, what it sounded like, when it started.
Time‑synchronised networking between arrays is in development on our next controller board.

Where it pays off

Built for plants that cannot afford surprises

Most failures are audible before they are visible. An acoustic camera turns that into a picture your maintenance team can act on.

Compressed air & gas leaks

Leaks hiss at frequencies the array pinpoints from across the hall, while the line keeps running.

Rotating machinery

Bearings, gearboxes, pumps and fans change their sound long before they stop. The camera shows which one.

Electrical assets

Partial discharge and arcing in switchgear, insulators and transformers are audible before they are visible.

Valves, steam & HVAC

Passing valves, failed steam traps and chattering actuators each have a signature of their own.

Safety & abnormal events

Impacts, alarms and unexpected noises, located and logged together with a picture of where they happened.

Noise mapping

Find the dominant sources on a floor, rank them, and prove that the fix worked.

Back of the 60-microphone array board with the MEMS microphones

Hardware

We design the whole stack

From the microphone PCB to the FPGA logic to the model on the NPU. Owning every layer is what makes a real‑time acoustic camera fit in one small box.

60‑channel hexagonal MEMS array on a single board, with a camera at its centre.
FPGA front‑end that acquires all channels and computes the sound map in hardware.
Edge SoC with NPU for sound‑signature classification and H.264 streaming.
Next: time‑synchronised networks of cameras covering a whole plant (in development).

Team

Hardware people, end to end

Decades of FPGAs, embedded Linux, signal processing and edge AI, in one small team that ships hardware.

Camilo Soto

CEO

Computer scientist with 10+ years in FPGAs, embedded systems and computer vision. Brings complex algorithms to hardware for real‑world impact.

Andrés Calderón

CTO

Computer scientist with 20+ years building devices from microcontrollers to FPGAs and neural accelerators. Signal processing and AI inference on edge devices.

Carlos Camargo, PhD

CHO

Electrical engineer with 20+ years in embedded systems, embedded Linux and HW/SW co‑design. Bridges hardware and software into cohesive system architectures.

See the sound.
Find the fault first.