ARCHEO: A Dataset for Sound Event Detection in Areas of Touristic Interest

Abstract

In this work we introduce ARCHEO; a dataset for audio event recognition consisting of audio clips collected from areas of touristic interest. ARCHEO audio samples can be divided into two types based on the way they were collected. The first type is comprised of a large pool of field recordings obtained through field trips in touristic areas near the center of Athens, Greece. The second type consists of a smaller pool of audio clips extracted from YouTube videos recorded at popular touristic destinations. In both types of data the audio is recorded from a typical smartphone and no special post-processing is carried out. This means that models trained on ARCHEO can be used on 'raw'data collected in a practical and cheap manner. ARCHEO contains 13 labels taken from the Audio Set Ontology (ASO) which can be used to extract information about the general environment of a subject.We give different baseline methods that illustrate the utility of our proposed dataset. © 2020IEEE.