Scholarlink

Environmental sound classification (ESC) has significant applications in acoustic monitoring and smart environments. This paper presents a novel multi-channel approach using attention-enhanced EfficientNet with audio-specific optimizations. Our method transforms audio signals into three-channel representations combining mel-spectrograms with delta and delta-delta features, employing squeeze-and-excitation attention for discriminative feature learning. We achieve 92.0% accuracy on ESC-50 without external pretraining, outperforming existing methods by 3.3% points. Key contributions include: (1) novel multi-channel audio representation capturing spectral and temporal dynamics, (2) attention-enhanced EfficientNet-B2 architecture, (3) comprehensive training with advanced augmentation, (4) rigorous experimental validation with statistical analysis.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Multi-channel Environmental Sound Classification Using EfficientNet

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Apara Maity,
  • Ausaf Aslam,
  • Ruchika Malhotra

摘要

Environmental sound classification (ESC) has significant applications in acoustic monitoring and smart environments. This paper presents a novel multi-channel approach using attention-enhanced EfficientNet with audio-specific optimizations. Our method transforms audio signals into three-channel representations combining mel-spectrograms with delta and delta-delta features, employing squeeze-and-excitation attention for discriminative feature learning. We achieve 92.0% accuracy on ESC-50 without external pretraining, outperforming existing methods by 3.3% points. Key contributions include: (1) novel multi-channel audio representation capturing spectral and temporal dynamics, (2) attention-enhanced EfficientNet-B2 architecture, (3) comprehensive training with advanced augmentation, (4) rigorous experimental validation with statistical analysis.