This study applies feature aggregation techniques in sound signal classification and localization networks. Feature aggregation consolidates information from various scales, enhancing feature robustness and invariance, thus improving the model’s performance. The researchers propose a new architecture, the Scale Encoding Network (SEN), for more computationally efficient aggregation. The researchers evaluated the efficacy of feature aggregation by integrating computer vision feature aggregation sub-architectures into a sound source localization control architecture. The results suggest that models incorporating feature aggregations outperformed the control model in both sound signal classification and localization.
Publication date: 3 Nov 2023
Project Page: https://gitlab.com/dsim-lab/paper-codes/feature-aggregation-for-neural-networks
Paper: https://arxiv.org/pdf/2310.19063
