AUDIO EVENT ANALYSIS METHOD IN NETWORK-BASED AUDIO ANALYTICS SYSTEMS

Relevance. In the rapidly evolving field of network-based audio analytics systems, the detection and analysis of audio events play a crucial role across various applications, including security, healthcare, and entertainment. Subject. This paper examines a method for recognizing audio events in network-based audio analytics systems, including preprocessing, sound separation, and the creation of machine learning models for analyzing audio signals. Objective. The objective is to develop and improve integrated methods for analyzing audio signals in network-based audio analytics systems to enhance the accuracy, speed, and reliability of data analysis. Methods. The proposed approach uses a modified ResNet architecture for multi-event classification and a convolutional neural network for separating sound sources in multi-channel recordings. Results. The method achieves competitive results, comparable to baseline results in contemporary challenges like DCASE and demonstrates robust performance in noisy environments. Conclusions. The proposed method shows potential for improving the accuracy and reliability of audio event recognition in real-world scenarios, particularly in complex acoustic environments.