Individual recognition is the basis of all animal interactions, and this becomes especially important in species that have a dynamic social environment. Therefore, many gregarious animal species have evolved individually distinct signals. In this project we explore whether it is possible to automatically record, detect and classify individual vocalisation signatures to construct social networks of a free-living songbird, the zebra finches. Zebra finches are the best studied songbird in the laboratory, but we have only a rudimentary understanding of how, with whom, and why these birds vocalise in their natural environment. As individual song signatures of male zebra finches are classifiable by human observers, they offer an excellent study system to investigate the automation of the sound detection and identification process.

In this project we collect recordings of zebra finch vocalisations using automated sound recorders (Songmeters 3 and 4, Wildlife Acoustics) in their natural environment. With the use of detection and classification machine learning approaches, we aim to transform these extensive audio streams to meaningful information on which individual was where at which time. From this who/where/when-information we can then extrapolate a social network in the natural environment, allowing us to gain insight in the social organisation of this species.

If we can get this to work, this will be extremely exciting, as normally it is not feasible to follow birds in the natural environment without catching, marking, and (automatically) reobserving them in some way. Given that zebra finches (and individuals of many other species of potential interest) may only be around a particular area for a short time, due to e.g. nomadism, this proof-of-concept will constitute a relatively cheap alternative for tracking individuals of vocally active species.