Leaky Echo State Network for Audio Classification in Construction Sites

In this work, we propose a machine learning-based classification approach aiming at identifying real-world sounds recorded in construction sites. The proposed approach is based on the leaky version of the Echo State Network (ESN), and it has been tested on a real-world dataset composed of recordings of five vehicles and tools usually used in construction sites. The implemented leaky version of the ESN exploits different spectral features as input. After the description of the proposed approach, we provide some numerical results obtained on the recorded signals. The overall accuracy on the test set, after the integration of a majority voting approach, is up to 95.3%, comparable to other state-of-the-art machine learning methods demonstrating the effectiveness of the approach.