| Title |
Frequency Dynamic Convolutional Neural Network based on Multi-resolutional Self-attention for Sound Event Detection |
| Authors |
이찬규(Chankyu Lee) ; 박상욱(Sangwook Park) |
| DOI |
https://doi.org/10.5573/ieie.2024.61.10.150 |
| Keywords |
Sound event detection; Separable convolution; Multi-resolution residual convoution; Frequency dynamic convolution; Attention |
| Abstract |
Sound event detection, which aims to identify the type of sound and interval of each sound event, is a crucial technique of various monitoring systems. Recently, various approaches to improve the performance of sound event detection are introduced through the international competition on sound signal analysis (DCASE: Detection and Classification of Acoustic Scenes and Events). To further improve the performance of sound event detection, it is needed to extract sound features from sound signal effectively. This paper analyzes the performance of existing methods that can be applied to design of Convolutional Neural Networks for building a sound event detection model, and proposes a more effective approach. The proposed model records 39.691.29[%] in the class averaging event-based F1 score in the detection experiment based on the DCASE2020 task4 testbed, showing the best performance compared to other conventional approaches. |