Case Study

Next-Gen MilTech:
NLP Annotation for Acoustic Target Detection

Location:

Ukraine

Services:

Audio annotation

Case Background

This data annotation project involves a number of AI initiatives in the field of military technology, aka MilTech. Since 2022, the Label Your Data team has partnered with other Ukrainian AI companies and young startups with the shared goal of helping the Armed Forces of Ukraine to leverage AI for strategic advancements. Our next case is a specialized project called Zvook.

Let the Numbers Speak: The Project Summary

+700 audio hours
annotated

40 complexes in full
operation

Closing a 10% gap in air target
destruction with AI

Audio Annotation Challenge for Label Your Data

The Client turned to Label Your Data with the goal of finding a trusted data labeling partner to develop an ML system for acoustic air target detection. For the algorithm to accurately detect an air target amid numerous objects (making similar sounds) and alert a user at the station, it requires the right amount of well-annotated audio data.

Zvook was designed to address the difficulty of tracking cruise missiles, particularly in scenarios where conventional radar systems, designed for detecting such targets, are incapacitated or when the target is flying at low altitudes. Label Your Data has joined the mission.

Main challenge

The project required deep Natural Language Processing (NLP) expertise from our annotators to handle audio data correctly. So the task at hand was to annotate audio files to further use them for the Zvook’s model training. The goal was to teach the algorithm to detect the sound of the flying missile and distinguish it from similar sounds.

Delivering Quality Data for the Air Target Detection System Development

The Label Your Data team has collaborated with various AI companies and the Armed Forces of Ukraine on this specialized project. Zvook’s advanced hardware and software system was aimed to acoustically detect air targets like missiles, helicopters, drones, and jet fighters at low-to-medium altitudes. The goal was to enable the military to track and intercept distinct missile sounds effectively.

Our annotators were specifically trained for this project to maintain the highest level of security and be able to distinguish between different sounds related to air target detection.

Challenges

Solutions

Training the network with annotated data for effective sound-based target recognition

Manual data labeling to augment the dataset with the noise sound samples that match real missile sounds

Preventing from system’s failure in detection of air targets

Ensuring data accuracy and reliability by conducting a thorough QA

Assisting the Zvook’s model in categorizing air targets

Labeling audio files with corresponding labels (i.e., helicopters, rotorcraft, jet aircraft, cruise missiles, and drones)

Supporting the project’s military communication

Gaining valuable military expertise and a keen understanding of Ukraine’s critical air defense infrastructure through cooperation with the brigade.

Solutions

Manual data labeling to augment the dataset with the noise sound samples that match real missile sounds

Ensuring data accuracy and reliability by conducting a thorough QA

Labeling audio files with corresponding labels (i.e., helicopters, rotorcraft, jet aircraft, cruise missiles, and drones)

Gaining valuable military expertise and a keen understanding of Ukraine’s critical air defense infrastructure through cooperation with the brigade.

The Label Your Data team helped the project with hundreds of hours of data manual labeling, which is one of the key factors of successful ML algorithms used in the Zvook system.

Pavlo Tsiupka

Project member and CEO of i3 Engineering

The Zvook Project
Results

For this successful short-term project, we annotated around 700 hours of audio files, distinguishing missile sounds from other recorded sounds we received from different stations that are located across Ukraine (i.e., noises like wind, animals, and explosions). We worked diligently for about two months, with a team of four dedicated annotators.

Here’s what we achieved in close cooperation with other AI companies and Armed Forces of Ukraine:

Increased chances of destroying the missile

The cruise missile was detected only four hours after setting up the Zvook system

Better data acquisition in air defense for smarter decisions in countering specific targets

Solved the lack of information on missile paths

Enhanced predictive capabilities of the model through expert audio annotation

Improved radar network

Performed additional surveillance of the air targets in “blind spots”

Created novel systems to fill the gaps in areas lacking radar installations

Improved the reliability of the Zvook’s system

Working with other AI companies and the Armed Forces, we helped enhance the system’s capabilities in detecting and neutralizing potential threats in real-time

Our contributions to manual data labeling played a crucial role in developing the Zvook system’s algorithms for sound-based air target detection.

Optimize your NLP project with
expert data annotation!

CONTACT NOW