.png){kind=icon}
返回
Multi-Level Transfer Learning using Incremental Granularities for environmental sound classification and detection
.png){kind=icon}
PRE
AI
DOI：10.1016/j.asoc.2024.112619.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
As sound recognition and classification models become more complex, their performance improves, but this comes with higher computational demands. This work addresses the application challenges in Environmental Sound Classification (ESC), where tasks often face the constraints of limited computational resources on edge devices and data scarcity. Leveraging the success of our previous Multi-Level Transfer Learning work in natural language processing and image recognition, we propose a Multi-Level Transfer Learning Using Incremental Granularities framework for ESC tasks. Our method utilizes audio features at multiple granular levels to provide varied learning characteristics, significantly enhancing the performance of small models without increasing model parameters too much. Experimental results show notable accuracy improvements across multiple models, particularly for smaller models. Specifically, our framework improved accuracy by up to 6.45% on the ESC-50 dataset and up to 8.6% on our custom chainsaw sound dataset, highlighting the advantages of transfer learning across different granularities.
Keyword:
Environmental sound classification and detection
Deep learning model
Multi-level transfer learning
期刊
IF:
6.6
论文数:
1.4W
被引数:
4.8W
机构
暂无机构信息
引用论文
Enlargement of rat submandibular salivary gland induced by single amputation of lower incisor teeth
PREAI
分享收藏A survey on deep learning tools dealing with data scarcity: definitions, challenges, solutions, tips, and applications关于处理数据稀缺的深度学习工具的调查: 定义，挑战，解决方案，技巧和应用.png)
JOURNAL OF BIG DATA
OAAI
JOURNAL OF BIG DATAIF6.4
OAAI分享收藏How to measure uncertainty in uncertainty sampling for active learning主动学习不确定度抽样中如何测量不确定度MACHINE LEARNING
OAAI
MACHINE LEARNINGIF2.9
OAAI分享收藏Attention based convolutional recurrent neural network for environmental sound classification基于注意力的卷积循环神经网络环境声音分类NEUROCOMPUTINGOAAI
NEUROCOMPUTINGIF6.5
OAAI分享收藏Multi-level transfer learning for improving the performance of deep neural networks: Theory and practice from the tasks of facial emotion recognition and named entity recognition用于提高深度神经网络性能的多级迁移学习: 来自面部情感识别和命名实体识别任务的理论和实践
PREAI
分享收藏
查看更多内容