Furnariidae species recognition using speech-related features and machine learning

Abstract

The automatic classification of calling bird species is important to achieve more exhaustive environmental monitoring and to manage natural resources. Bird vocalizations allow to recognise new species, their natural history and macro-systematic relations, while automatic systems can speed up and improve all the process. In this work, we use state-of-art features designed for speech and speaker state recognition to classify 25 species of Furnariidae family. Since Furnariidae species inhabit the Litoral Paranaense region of Argentina (South America), this work could promote further research on the topic and the implementation of in-situ monitoring systems. Our analysis includes two widely-known classification techniques: random forest an support vector machines. The results are promising, near 86%, and were validated in a cross-validation scheme.