Audio diarization for biodiversity monitoring

Abstract

Biodiversity monitoring is important in assessing the state of an ecosystem and determining if conservation actions are required. This is particularly important when conservation resources are scarce. However, traditional methods of biodiversity monitoring are labour intensive and cannot be applied in every ecosystem where there is need. In order to expand the application of biodiversity monitoring, there is need to automate this important task. In this work we present an application of audio diarization methods for biodiversity monitoring and show how these methods can be used to measure the abundance of indicator taxa in areas of interest. The use of audio recordings has the potential to reduce the time and effort spent in biodiversity monitoring. The experiments are performed on a freely available dataset of bird song recordings with the birds serving as indicator taxa in the ecosystem of interest. We are able to estimate the number of bird species in the recordings and this information can be used to estimate the species richness in an ecosystem.