Environment Sound Classification (ESC) with Choquet Integral Fusion

Document Type

Conference Proceeding

Publication Date

1-24-2022

Department

Department of Computer Science; Department of Mechanical Engineering-Engineering Mechanics

Abstract

The Choquet integral (ChI) plays an important role in the area of aggregating sensors and information. One of the defining advantages of the ChI, compared to other types of aggregations, is that it takes into account how variables interact with one another, which it does by means of what is called a capacity or fuzzy measure. The fuzzy measure captures the relative value, or worth, of different subsets of information sources taken together to make an inference. For data-driven problems, i.e., machine learning, the fuzzy measure comprises the parameters learned for using the ChI. In this work, we apply data-driven ChI decision-level fusion to the problem of classifying sound events from clips of audio. Three benchmark sound classification data sets are utilized: ESC-10, ESC-50, and UrbanSound8K. Six leading classification algorithms-deep networks and transformers-are used as the decision sources. The ChI fusion shows significant gains in accuracy for all three benchmarks.

Publication Title

2021 IEEE Symposium Series on Computational Intelligence, SSCI 2021 - Proceedings

ISBN

9781728190488

Share

COinS