Sound-based Multi-Equipment Activity Recognition

Published: 3 September 2021| Version 1 | DOI: 10.17632/7vsbyfrhc4.1
Behnam Sherafat


This dataset is recorded on real noisy construction job sites. It consists of two main sound data: 1) single-equipment and 2) multi-equipment. It consists of five case studies. The first three case studies consist of single-equipment sound for two types of heavy equipment. The last two case studies consist of both single-equipment sound data and real-world multi-equipment sound mixture. The purpose of this dataset is to train a model using single-equipment sound data and test it on multi-equipment sound mixtures for both synthetic and real-world sound mixtures. This dataset is recorded using an off-the-shelf Zoom H1 digital handy recorder, which is a single-channel microphone. Because the dataset is recorded using a single-channel microphone, the process of the multi-label sound classifier is considered a software-based method. On the other hand, there are many datasets that are recorded using microphone arrays. The process of the multi-label sound classifier for these datasets is considered a hardware-based method.