Deep Echo State Network for Environmental Sound Classification

Authors

• Michele Scarpiniti, Sapienza Università di Roma
• Sofia Perticarà, Sapienza Università di Roma
• Yong Cheol Lee, LSU College of Engineering
• Aurelio Uncini, Sapienza Università di Roma

Document Type

Conference Proceeding

Publication Date

1-1-2025

Abstract

In this paper, we propose a deep learning based approach for the environmental sound classification (ESC). The proposed approach is based on a deep version of the Echo State Network (DeepESN) and it has been tested on the well-known ESC-10 dataset and on some real-world recordings of vehicles used in construction sites (CS dataset). The implemented DeepESN exploits different classifiers working on the extracted hidden features. The overall accuracy on the test set is up to 77.8% for ESC-10 and 99.6% for the CS, comparable to other state-of-the-art machine learning methods and demonstrating the effectiveness of the approach.

Publication Source (Journal or Book title)

Smart Innovation Systems and Technologies

First Page

71

Last Page

82

Recommended Citation

Scarpiniti, M., Perticarà, S., Lee, Y., & Uncini, A. (2025). Deep Echo State Network for Environmental Sound Classification. Smart Innovation Systems and Technologies, 428, 71-82. https://doi.org/10.1007/978-981-96-0994-9_7