This Ukrainian start-up detects Russian cruise missiles in trouble because of a prisoner on hunger strike – Meloni settles with a letter

A menacing roar, like that of a low-flying fighter jet, followed by a singing hum. The background noise of a low-flying Russian cruise missile sounds familiar and yet strange at the same time.

Countless cell phone videos show the eerie scenes of passing Russian missiles on their way to their Ukrainian targets. Unlike ballistic missiles, many cruise missiles fly very low for most of their journey, often at an altitude of 20 to 100 meters and usually below the speed of sound.

The low altitude makes it difficult, especially for radar systems, to detect them in time and to take appropriate countermeasures. There are also a number of regions in Ukraine that are not under radar surveillance.

The Ukrainian air defense is now receiving increasing support from the Zvook AI project, a system that, in addition to cruise missiles, can also recognize and locate low-flying helicopters, drones and fighter jets by their engine sounds.

There’s an old technique behind this: during World War II, people on sound posts tried to hear approaching V2 rockets in time, says Pavlo Tsyupka, one of the co-founders of Zvook, the Ukrainian online magazine “Pravda”. With the help of special mirrors, the sounds from the listening direction were amplified.

Zvook is the digital version of such listening posts: Sensitive microphones are combined with appropriate mirrors and several such devices are then connected to form so-called microphone arrays. In this way, the flight direction of an intercepted flying object can be determined.

But for the recognition to work quickly and automatically, a special AI system had to be developed and trained for the desired sounds. The microphone arrays now continuously send the recorded signals via a secure data network to a decentralized computer network, which takes over the analysis and, in the event of a hit, forwards the results to the responsible units of the Ukrainian army.

Initially, the detection rate was only 50 percent

Technically, this is a big challenge. According to the creators, about 99.9 percent of the sounds are uninteresting from the start. The remaining 0.1 percent should be filtered out of the signal and analyzed.

AI systems can perform such a task quickly and accurately. But they also need to be trained beforehand – and that requires as many recordings as possible of the sounds to be recognized. Initially only the sounds from mobile phone recordings were available for this – however, they did not have the same sound profile as the noise recorded by the Zwook microphones and the quality of the mobile phone microphones was also poor.

At the start of development, the recognition rate was only about 50 percent – far too low for semi-flawless operation. Only after the makers won the support of the Ukrainian army and were allowed to install their systems on the transmission towers of the network operators, did they manage to collect enough audio material from Russian cruise missiles for weeks and months.

Zvol as a supplement with an important advantage

These had to be recognized and isolated by humans first. But the growing pool of suitable shots finally enabled effective training of the AI. The system is now very reliable and will only get better the longer it works.

Zvook is therefore by no means a replacement for radar systems, but an important addition. It is relatively quick and easy to manufacture – and unlike radar, which actively emits radiation and can thus become the target of enemy attacks, Zvook microphones operate purely passively.

Currently, a mid-double-digit number of the systems are in use and about 600 are needed. reach their target, but can be intercepted in advance.