Рақамли технологияларнинг назарий ва амалий масалалари 9-том 3-нөмір (2026) · 53-74-беттер
Analysis of methods and approaches for sound event detection in emergency situations
Набиева, Д.Т., Юлдашева, У.Х.
Аңдатпа
This paper presents a comprehensive review of existing methods and approaches for Sound Event Detection (SED) in environmental acoustic monitoring. The theoretical foundations of SED, acoustic feature extraction techniques (MFCC, Mel-spectrogram, DCT-spectrogram, and cochleogram), traditional machine learning methods (GMM, HMM, and SVM), and modern deep learning architectures (CNN, RNN/LSTM, CRNN, and Transformer) are comprehensively analyzed. The review shows that between 2000 and 2012, manually engineered features combined with conventional machine learning methods achieved detection accuracies of 65–75%, whereas deep learning approaches introduced after 2012 increased performance to 85–96%. In particular, Convolutional Recurrent Neural Networks (CRNNs) achieve accuracies exceeding 90% by effectively modeling local spectral patterns and temporal dependencies while benefiting from ensemble learning techniques. The paper also analyzes widely used benchmark datasets (ESC-50, UrbanSound8K, AudioSet, and FSD50K) and highlights the need for developing specialized datasets for emergency sound detection in residential environments. Furthermore, practical challenges such as noise robustness, real-time inference, and edge computing are discussed, together with promising research directions including Transformer-based architectures, self-supervised learning, and multimodal fusion. The presented review provides a valuable reference for researchers and practitioners working in the field of sound event detection.
обнаружение звуковых событийглубокие нейронные сетиCRNNакустический мониторингумные городачрезвычайные ситуациимашинное обучениеsound event detectiondeep neural networksCRNN
Метадеректер дереккөзі: журналдың OAI-PMH архиві · Sindex толық мәтінді сақтамайды, дереккөзге сілтеме береді.