Multi-Instance Multi-Label Class Discovery: A Computational Approach for Assessing Bird Biodiversity

Forrest Briggs, Xiaoli Z. Fern, Raviv Raich, Matthew Betts

Abstract: We study the problem of analyzing a large volume of bio-acoustic data collected in-situ with the goal of assessing the biodiversity of bird species at the data collection site. We are interested in the class discovery problem for this setting. Specifically, given a large collection of audio recordings containing bird and other sounds, we aim to automatically select a fixed size subset of the recordings for human expert labeling such that the maximum number of species/classes is discovered. We employ a multi-instance multi-label representation to address multiple simultaneously vocalizing birds with sounds that overlap in time, and propose new algorithms for species/class discovery using this representation. In a comparative study, we show that the proposed methods discover more species/classes than current state-of-the-art in a real world dataset of 92,095 ten-second recordings collected in field conditions.