음향 이벤트 검출 모델에 적용된 Pooling 방법에 따른 성능 변화에 관한 고찰

다국어 초록 (Multilingual Abstract)

This study explored the effectiveness of combining max pooling and average pooling techniques to enhance the performance of acoustic event recognition systems. Through experiments conducted using the DESED dataset, it was found that the model employing max pooling at the initial layer significantly droped recognition performance compared to the baseline model. This finding suggests that there may be an optimized pooling technique suitable for each layer within the model. Future work will focus on further enhancing performance by developing and applying layer-specific optimized pooling techniques.