Acoustic drone detection for critical infrastructure: evaluating Sound Shield and the plan to wire power grids

Neuron Soundware's Sound Shield uses low-cost microphone sensors and on-device AI to detect drones by their engine noise, positioning the system as a passive, affordable alternative to radar for protecting electrical grids. The company plans to deploy sensors on transformer stations for dual-use monitoring of infrastructure health and airspace threats, but key performance claims have not been independently validated.

What happened

Czech startup Neuron Soundware has built Sound Shield, an AI system that identifies drones by the sound of their engines using microphone sensors that cost between €100 and €150 each. The system deploys small sensors called nEdge Minis, each consuming 1 watt of power, that listen continuously for drone signatures. Audio is processed on-device by Nvidia Jetson modules running neural networks that match incoming sound against a library of known drone acoustic profiles. When a threat is detected, the system reports the drone's estimated speed, altitude, and direction to a centralized command platform.

The company, which has spent the past decade using AI to monitor industrial machinery for clients including Airbus, Siemens, and BMW, is now applying the same acoustic analysis technology to airspace defense. To date, Neuron Soundware has raised approximately €7.4 million from investors including Inven Capital, J&T Ventures, and Lead Ventures, and received €7 million from the European Innovation Council. The company has over 130 industrial installations across four continents monitoring machinery acoustically.

Why AI builders should care

Sound Shield's approach is notable for its dual-use deployment model. Rather than asking governments to fund a standalone drone detection network, Neuron Soundware is proposing to piggyback on infrastructure that already needs acoustic monitoring. Founder Pavel Konečný told TNW that transformer station sensors could “continuously monitor the health of the transformer itself and other critical components of the distribution network, detecting internal discharges, oil leaks, or other operational anomalies” while “their microphones listen to the sky.”

For AI builders working on edge inference or sensor fusion, this case illustrates how existing acoustic monitoring infrastructure can be repurposed for security applications. The sensors are passive - they emit no signal that an adversary could detect or jam - which is a meaningful advantage over active radar. The system also demonstrates a practical deployment of Nvidia Jetson for real-time on-device inference in a low-power, constrained environment.

Practical implications

The cost argument is straightforward: modern radar systems capable of detecting small drones cost orders of magnitude more than a network of nEdge Minis. The counter-drone market is expected to more than triple from roughly $6.6 billion in 2025 to $20 billion by 2030, and acoustic detection is positioning itself as a complementary layer to radar and radio-frequency detection.

However, the practical performance of acoustic drone detection has well-documented limitations. Most acoustic systems are effective to roughly 300-500 metres under favorable conditions, with performance degrading substantially in wind, rain, or noisy urban environments. Ambient noise from traffic, wildlife, and industrial equipment can produce false positives. Newer drone models are being designed with quieter motors that reduce the acoustic signature available for detection. Neuron Soundware claims its nEdge PRO computing module can aggregate data from sensors within a 20-kilometre radius, but independent testing of that range claim has not been published.

Caveats

Most claims about Sound Shield come from company materials and promotional coverage; there is no published independent validation of its detection range, reliability, or false-positive rates in real-world conditions. The jump from listening to pumps and turbines in industrial settings to tracking hostile drones in contested airspace may not be as straightforward as the company suggests. The system's effectiveness will depend heavily on environmental factors, and the 20 km aggregation radius for nEdge PRO remains an unverified claim.

FAQs

What is Sound Shield and how does it detect drones by sound?

Sound Shield is an acoustic drone detection system from Neuron Soundware that uses microphone sensors costing €100-€150 each to identify drone engine signatures in real time. Detection relies on comparing incoming audio to a library of known drone acoustic profiles, with processing performed on Nvidia Jetson devices.

What are the typical sensor costs for acoustic drone detection systems?

Reported sensor costs for Sound Shield are €100-€150 per sensor according to company promotional material. Broader pricing for rival acoustic systems is not detailed in the provided sources.

How can acoustic drone detection complement radar and RF-based counter-drone systems?

Sound Shield is positioned as a complementary layer to radar and RF detection, not a replacement. The passive acoustic approach enables cheaper deployment by piggybacking on existing infrastructure, while radar and RF systems cover different detection modalities.

Where are acoustic drone detection systems like Sound Shield typically deployed (e.g., transformer stations, electrical grids)?

The dual-use pitch targets deployment on electrical transformer stations to monitor infrastructure health while sensing airspace threats. Company statements emphasize transformer-station applications and broader critical infrastructure monitoring.

What processing hardware does the Sound Shield use for on-device AI inference?

On-device AI inference runs on Nvidia Jetson modules that process audio against drone acoustic profiles in real time.

What validation or range data exist for the effectiveness of acoustic drone detection systems?

Independent validation of the claimed 20 km aggregation radius for the nEdge PRO hardware is not published in the provided sources. Practical range data cited by the company are 300-500 metres under favorable conditions, with performance degrading in wind, rain, or urban noise. No independent third-party tests are available.