A Visualization Framework for Feature Investigation in Soundscape Recordings

Abstract

Studies in soundscape ecology can generate large volumes of audio recordings collected over extensive time intervals. Extracting information from such data is challenging and time demanding. Important tasks, in this context, are to identify occurrences of acoustic events of interest and find out which combination of audio features are suitable for characterizing specific events or describing a particular soundscape. Researchers in soundscape ecology have been investigating approaches to accomplish such tasks effectively, and there is a demand for tools capable of assisting analysts in investigating large databases of ecological recordings. In this paper we describe a visualization framework for this purpose. The system includes multiple functionalities for soundscape analysis, comprising audio feature extraction, identification of relevant acoustic events by means of visualizations associated with audio playbacks, and event characterization by means of subspace feature analysis, also assisted by visualizations. The system implements a user-driven iterative pipeline that gives domain experts means to search for, identify and characterize acoustic events, gathering insight on which features better describe them and their originating soundscape.