he is fighting we will ask our guest today, this is maryan sulym, a member of the sound project and a serviceman of the 125th separate brigade of the territorial defense of the city of lviv, mr. maryan, i congratulate you. so, the first question is how exactly was this sound project born, what does he imagine himself to be , how does it work in order to people had the impression that what exactly we are going to discuss today was the back of it. for the implementation of this project, one volunteer from lviv, the father, as they are called now, was the main goal of protecting lviv, including the opportunity to protect the whole of ukraine from air attacks of our enemy, the russians. he came to the 125th separate brigade of territorial defense of lviv with such a request and received

the full support of our generals. artur gorbenko did everything possible for this project to be born and begin to develop . described in detail in principle, but you already wrote it in great detail, you can say that it detects air targets, this is its main purpose. it was created by an impulse, of course there was a problem that immediately became clear to everyone. at the beginning of the war, it was very difficult for us to track the cruise missiles with which we were attacked during, in principle, the first days of the war, when many classic radar systems for detecting such targets were destroyed. second, they generally used very

effective tactics on e-e using flight at low altitudes where classic radar systems already show their low lower efficiency due to the curvature of the earth so m-m so acoustic detection method it has a lot of advantages compared to the classic ones, first of all, this is what it actually says that there is a certain device that receives the sound produced by an enemy missile, this device is located somewhere, this device will receive a sound signal, it is processed and it is possible to clearly find out what type of enemy object is flying by this device at this time, so in general, in relation to the type, we are now dividing artificial intelligence, it can divide air targets into hm , such categories as propellers, helicopters, airplanes

propeller jets or cruise missiles or drones. that is, this is a classification into three types. we have already gone much further than, in principle, such systems have been able to do for the last 80 years and 100 years, because such systems have existed for a very long time, even in the thousand even when the first radars appeared in the 1930s, acoustic direction finders, as they were called then, already existed since the 1880s, there were sometimes such people sitting with such big ears listening to the direction of the sound. the history of the first world war, because, as i said, from the 1930s, they began to gain momentum, and now also became classic radio location

methods for detecting aerial targets, and from the 1880s, they were widely used on ships in order to monitor the water area in tuman -e and, accordingly, you could hear the sounds of the engines of the engines in the first world war, they also detected air targets eh, also an interesting fact would probably be to note that the first airplanes that appeared in the first world war , they were, in principle, radio permeable, they were not they reflected, that is, they did not see radars, radar did not see them in principle, that is why when the first radars appeared in the 30s and they were already inefficient and acoustic sensors, they were still used for a very long time, even during the second world war, we basically used from that of all the experience that i have just listed, we used only the fact of the selection

of materials, because when they began to implement them, we quickly realized that metal is not the best material for the implementation of such an acoustic direction finder it introduces a lot of um interference changes to the signal itself and the sound itself and that's why we already had a certain idea how we should build this system before the start of the project, that is, in fact, we are now talking about the fact that we have hard in this the receiver itself, let's say so so is the software that processes the signal that comes into this conditional receiver, yes, that is, if you can tell me briefly, now on our screen, the viewers can see what your receivers actually look like, that is, how it is made, where it is located, first exactly the prohar and then a little prosoft here are the details to make it interesting. the main part of the sound sensor of our project is the very mirror that helps to focus the sound on the

microphone. in this way, we achieve a longer detection range, because in general, we can talk in more detail later about different systems that are similar to ours, but they everyone has their pluses and minuses and we we had to make a decision in which direction we are moving and which problems we close, accordingly, our implementation in this form resembles as much as possible constructor, but still we hmm if we stand on the fact that we need to have a longer detection range , we achieve this at the expense of a mirror, it is a parabolic shape, inside there is a single microphone to ensure the detection of the angle for the purpose of this is not enough, of course, but in principle it is not we need all implementations, which one do we put

, and where is this microphone, or shall we say this is a political mirror, where is it placed, we install it at a height of 10-12 m, which also allows us to increase the detection range of any telecommunications towers, we can put them, all that is required from the requirements is the availability of power, we can even provide our own additional uninterrupted e-e, we saw that these parabolic sound signal receivers can be placed on all our telecommunication towers, and this actually creates such an information network of detection in enemy facilities, yes, how many such samples are already standing and working in ukraine, we already have more than 40 such stations, closer to 50, and we also plan to double the number in the near future of these stations, it may take us, for sure, the next month, maybe a little more than two

months, we will work on it for the time being. so far , we cover more than 40 of the 200 km borders with them. that is, the network is already working there, with a width of 200 km , right here the sound field actually already exists and works yes yes we are trying to build the location of the station with the so-called boundaries, this allows us to see the progress if the flight of the target over the territory, that is, there is no content. especially we do not see covering the entire territory of the state in order to owning the air situation is enough to have them as borders and when crossing the border even when changing the course of the air target hm cruise missiles can do many times in one political program all the same the situation is visible and

understandable for our troops now then we will move on to the software a little, that is understood the most, this is a critical component of this complex, because it is actually necessary to somehow guarantee and process sound signals without error in order to clearly understand what exactly is flying by and by and by along ukraine and arises the question is, what did you use for this ? artificial intelligence, machine learning technologies, that is, as i understand it, you started working with sound processing, and then before this operative. and please tell me what other teams are involved in this, because it actually looks like it is a multi-component work where in fact, different skills of the participants are needed to really ensure an effective result, so definitely er-er this system could work without artificial intelligence, but artificial intelligence allows er-er to- make it fully automated, i.e. the activation of m.m. they are transmitted to m.m. to the operator’s workplace

in automatic mode and there is no point in listening to each sensor, as it was in the implementations in the second world war, precisely over the artificial intelligence system, we have many teams that are involved in this project specifically , a team of employees of the respicher company is working on the artificial intelligence model, they do it in their free time, they took it upon themselves, you can say i probably think 30% of this project, well, it is an artificial intelligence model, no we did not have enough resources to listen to all these sensors in manual mode. but this was just the beginning before we were able to build an effective model, in principle we had to collect all the signals manually, you can understand the current state of affairs on this war, if even

artificial intelligence is on our side, russia is coping with an extraordinary task, because linguistically, sound and noise are present in our everyday life, yes . that is, we live in sounds and this means that it is better to isolate what from this field of different sounds that is really related to the enemy object. that is , as i understand it, you had to collect various sounds that are related to missiles or drones, then isolate these sounds in different conditions and create a database of these sounds to put them in the computer in order to relatively speaking, against the background of this informational sound noise, single out exactly what is related to enemy

targets . which you can clearly see on the spectrogram when the sound is decomposed through the fourier transform into the spectrogram. are very strongly attenuated at distances of more than 500 literally meters, and already when we installed the sensors and air targets had a chance to pass between them, because it is impossible to cover the territory so tightly to put it at 500 every 500 m in principle, the road to nowhere, we immediately gave up on it, that is why we started to work on increasing the range at distances greater than 500 m e-e, most of the high times are already fading and

from there to two from zero to 2,000 hz we are already monitoring by mathematical methods e by level, that is, by signal strength of course, no span can happen with e-e environmental signal levels, so it will increase if we see any span, respectively, our first factor is when reaching at when increasing the signal level we are starting to send these spectrograms to medical artificial intelligence already for analysis, it works as a model. to be honest, it is very difficult to say, this is the so-called black box, there are processes that are in principle unknown to us, they are due to the fact that we

were able to collect these samples of these flights, the model was already able to distinguish er, all future flights in such and such spectrograms where it is no longer impossible for the human eye to see it, that is, now we can talk about the fact that we are guaranteed to be there from here, here is higher than 50%, maybe 80 percent we can clearly determine that it is the sound of a cruise missile, and on the width of the front there, roughly speaking, 200 km and in depth, i don't know, maybe now i can ask you what the detection depth is, we clearly know that a cruise missile is flying, and this is the basis for this the signal was transmitted to the air defense system to make an additional decision already on the destruction of this object, this is how it looks for cruise missiles and for aircraft .

from real flights in different weather conditions in different variations there in different directions and it really works for cruise missiles and the second question is for sure i will not deceive anyone. with drones, the situation is more complicated, that is, it is difficult to distinguish them from a moped. in principle, we believe that this can be done effectively, but unfortunately no one, since no one has a sufficient number of records, it is very difficult to get them, in principle we ask everyone we can reach to provide us with such recordings, but this is also a difficult technical task, the recordings are made on mobile phones, they are very suitable, they are very distorted, they are cut, in general , mobile phones are not suitable for training such models of artificial intelligence, they are not there , they are designed for human voices, my microphones in

mobile phones are already designed for transmission human voice and nothing else. that is, it is precisely the problem of forming an acoustic database for analysis, that is, an adequate database. that is, we can say that if we collected on cruise missiles just a high level of detection adequacy is ensured, i.e. just go to the internet and download records to get rockets or even if they are not very productive, it is used, it is called pre-training. that is , you can feed m.m. models of artificial intelligence, dear, such data will suit it in favor of such a percentage as we achieved as a result of training covered missiles, this does not look at how long it took for artificial intelligence to learn to adequately distinguish flight of a cruise missile in the air space in general, for sure, it is possible to say how long this task

should last in terms of time, it is definitely not possible, it all depends on the conditions in which our team was located, well, the main difficulties are that i can convey to you the main difficulties, such as if you have one sensor and it records the ether is permissible you have 24 hours of this environmental ether around this sensor er to find a flight in it god forbid if this flight or god forbid that this does not happen so that these flights do not happen at all was, but if in 24 hours this arrival will be 10-15 seconds and a person should listen to it, a person should hear it, identify it, cut it and save it. this was the main task of 40 sensors when we installed uh, hmm, it was necessary to plant 40 people so that they listen for 24 hours ether, precisely due to this mathematical detection of the level of signal activation, we achieved the fact that hmm, in general, flights and many other things were already

collected in automatic mode, and it took us, accordingly, your question four months only after four months when in all samples were collected in automatic mode, including far from cruise missiles and manually sorting them, we were able to achieve that in automatic mode, the following samples were already collected for us. i carefully read your previous interview and i will see some paradoxical things there that they are the most important during preparation adequate databases were hindered by cows, bees, and there, er, and other creatures that, conventionally speaking, can pass closely between the sound sensor and the spectrum in which they emit sounds is very close to, conventionally speaking, the flight of the rocket. i read it all. if so, how did you get out of this trap? well, for sure, after all, the mooing of a cow cannot be said to be close to a cruise missile on the spectrometer

. more likely, the reason was to teach the model to ignore them. it is necessary to collect and say, this is necessary, this is a rocket, yes, yes, it is necessary to find them. well, it was very difficult to collect records of cow records, but when they happened, we had to listen to them and add them to the exclusion, in principle, we did the same with mopeds and chainsaws, that is, in fact the database on mopeds based on sounds is currently being worked on to really distinguish a chess from a moped, despite the fact that to the human ear it seems to be very similar, that is, but the nuance that must be introduced in if this problem has a solution is very easy , in fact, if you put the engine in laboratory conditions shaheda is alive. hmmm, record it with a normal number of microphones, sound studio equipment, it can be done more than us,

if there are colleagues who are participants in this project, they are engaged in exactly this kind of work, they advise us on microphones on precisely technical the component eh of exactly how it works, how the microphone works, how it all happens at this level and they achieve in their daily work, they achieve such results that they can determine the degree of wear of the rotor by the sound of work eh the fields are not in the conditions of the wind, rain is why our task is more difficult. and when they talked about the fact that there are analogues, in what ways are these analogues from e-e close to your solutions and different in what is happening, let's say conditionally, there is a conversation about competition between projects that are directed at detection of enemy objects precisely by the sound

segment, well, this is very, very difficult, a very difficult topic , the main thing is that you need to understand that there is probably no single correct approach to the determination of the detection of air targets in particular. that is, we concentrated on our how to do it, the main approach is this or we we put many sensors all over ukraine and one of them will work with a high probability and during, for example, while an air target flies there for 20 or 30 or 50 km, one of these sensors with a high probability of catches or we do a different approach. that is, i am now telling you about other approaches that could be used. in general, we can absolutely act in the same way. it is as complex as the great designer. we can give up the antenna. then we will not take the directivity pattern that the antenna has, the mirror

, but will to have 360°, uh, detection, but it has its disadvantages, it has a shorter range, and such a microphone will not be protected from the external environment. that is, we have to create other reception modules, yes, we can do it, we can implement it, but having built such a network, we are afraid that we may help the state there for a few months, and then we will have to hand over all the difficulties of maintaining such a system to someone. imagine, for example, we can see that in order to close the main lines of flight of enemy missiles in the area of ​​600 of our stations, that in order to cover the main territories of ukraine along the perimeter and form a reliable field, yes, that is, now 600

sensors for the whole of ukraine, yes, that is, that is how we see such an approach, if we put not 600 of them, but put there are one and a half, two, three, four, five thousand of them, we will increase the density, we will increase the density, but who will service them, who will be able to service such a number of sensors? well, theoretically, if we put these sensors on towers, yes, that is, in particular, telecommunications can survive, in principle, it is possible to create such a symbiosis where in fact, the emotional operator undertakes, among other things, the obligation to service these sound-receiving devices and to put it in the task of the local administrations of the military or others so that it is such an integral project for your specialists and eh, the structures that take care of telecommunications is the same. it is possible to combine it, so we have always passed on such proposals, but in the conditions of war and in the conditions of the fact that from the beginning

, after all, we somehow managed to restore our classic radar system for detecting air targets, in general, the system of acoustic detection is not much, it is already the main one, it is even complementary, it should be complementary. and that is exactly the question. how many resources do you want to spend on a system that only supplements your basic radar classic ? therefore, we need to keep a balance between the amount of resources that we spend on this system and the advantages that we receive in our defense of our state, i.e. now we have an experimental area there, a 200 km wide patch of land, and on the basis of which we are starting to develop new target detection algorithms the main thing is dapaset, we created this uh, this is this milestone, first of all, in order to

teach the model of artificial intelligence to perform its work qualitatively. financial directly from your team , there is no participation of the state or local administrations in this so far. well, if you would call this a startup, it basically appeared as an initiative to build on the hmm roof, in the end it is called the construction of such a system from which you can come with a folder of paper where it is described and show that it works, please take it away, we will hand it over to you and implement it, because we still love military developments, they are made if, after all, from the side of the state we got into it, you can say for six months we went and told stories money and brains let's say that the efforts that will be directed specifically to the creation of this project, all that we have achieved so far from these activities will now be a test in kyiv of such

systems of ours, and also in principle, and well, we do not call them competitors, i just tried to explain that these are just different approaches that can be used and they should complement each other, they do not even exclude each other from the map of ukraine, well, now we are not at the stage where, by the way, the implementation of such new developments is essential simplified, i.e. now the test cycles have been shortened. and the military told me that i was approved by your company. if the product works, you take this product, prove that it is really effective, and they take it to support the armed forces, and on the basis of this , some conclusions are drawn up regarding the future fate of one or another project, as i see it, in fact, with your project, we have reached the stage of pressing the military, what, guys, look , it works. yes, we have come precisely for this stage . these tests will be carried out during in the next month and a half, we will

recommend anyone to offer any help from our project, whether it will be a network, it will be implemented by the state, or we will transfer these works, some of our ideas so that this system is developed from scratch, taking into account our experience that we already have until now, on all the problems we faced, which we overcame. hmm, we'll see how it will show itself by the end of the year. well, it's better now, as i understand it, and the main thing is that the person who supported it, as i understand it, was like the leadership of territorial defense, how can you say that in the alliance with the lviv military administration, it is clear that it can work. they trusted and you did it and it is now working, that is, yes, the local government believed in this project, you can

say from the first day the city government did everything to ensure that we they started working in that chaos because the project was born almost immediately, at first there was a war in the head, hmm, and after that, the implementation began already already in april, the first stations were already standing , you didn't tell me the depth of 200 km - it's with us along the front, how far can we hear an enemy missile of the khas1 type, conditional, they say, uh, they are standing, as i said, they are standing in lines. each of the lines can have a detection range, which we estimate to be from five to seven kilometers, we have no way to order a flight such a cruise missile in front of our sensor, ideally according to the directional pattern in the spa center, so i hope that during these laboratory tests that will take place, we will be able to find out with an accuracy of up to a meter what our real

range is, because take the x1 and they will really provide real sounds that correspond to these rockets i think that any jet engine would be enough in them the nature of the sound they emit is noise -like this is the main principle that the artificial intelligence model uses for identification i.e. if the level of this sound is noise-like higher than the ambient noise level of the road wind field that is, we we have to detect sound and if the sensor is at a height of 10-15 m, yes, that is why they say that there is a risk, relatively speaking, of sound signal reflection, yes , that is, if we simply detect sound, this value no, but if we want to know the direction of movement of a sound object, is it possible to implement it in your project, how does the enemy object move? hmm, first of all, you said for m-m for reflection. that is

, it is also one of the main factors why we chose this one approach to the implementation of such a project, since we have a mirror and a parabolic antenna and we do not have a 360 a 100° directivity diagram. thus, we can raise it to the horizon and we can cut off what is happening from the ground and from the noise, for example, of the road that is near of our sensor because we are essentially listening to the sky, and everything according to the directional pattern of our mirror it is weakened by approximately two times according to our calculations, we can even improve this indicator. that is, we can make a system that will listen only to the sky and this it will not, it will still help to increase the detection range, and the second question, which one are you

directly ?