On-Bird Sound Recordings: Automatic Acoustic Recognition of Activities and Contexts

Dan Stowell,Emmanouil Benetos,Lisa F Gill

doi:10.1109/taslp.2017.2690565

Abstract

We introduce a novel approach to studying animal behavior and the context in which it occurs, through the use of microphone backpacks carried on the backs of individual free-flying birds. These sensors are increasingly used by animal behavior researchers to study individual vocalizations of freely behaving animals, even in the field. However, such devices may record more than an animal's vocal behavior, and have the potential to be used for investigating specific activities movement and context background within which vocalizations occur. To facilitate this approach, we investigate the automatic annotation of such recordings through two different sound scene analysis paradigms: A scene-classification method using feature learning, and an event-detection method using probabilistic latent component analysis. We analyze recordings made with Eurasian jackdaws Corvus monedula in both captive and field settings. Results are comparable with the state of the art in sound scene analysis; we find that the current recognition quality level enables scalable automatic annotation of audio logger data, given partial annotation, but also find that individual differences between animals and/or their backpacks limit the generalization from one individual to another. we consider the interrelation of “scenes” and “events” in this particular task, and issues of temporal resolution.

Highlights

Studying the behaviour of animals in real time and in their natural environments is becoming more and more feasible through the use of animal-borne loggers or other remote sensing technology [1]
We evaluated each of our systems in two configurations: the classifier-based system with unbalanced or balanced class-weighting for training; and the probabilistic latent component analysis (PLCA) system with mean- or maximum-based temporal downsampling
The X-Y scheme in turn was similar to the EachCap scheme except that it pooled the training data across individuals. This pooling led to very similar F scores as EachCap, but to a marked difference in area under the ROC curve (AUC): judged by AUC, the pooling of training data seems to have led to better generalisation properties, for both of the recognition algorithms tested

Summary

Introduction

Studying the behaviour of animals in real time and in their natural environments is becoming more and more feasible through the use of animal-borne loggers or other remote sensing technology [1]. These technologies have provided insight into different aspects of physiology and behaviour, such as heartbeat [2] or migratory routes [3, 4], which in turn can help us understand basic mechanisms up to evolutionary drivers, as well as support decision-making processes in nature conservation or disease management.

Results

Discussion

Conclusion