Integration over song classification replicates: song variant analysis in the hihi

Area of Science:

• Bioacoustics
• Machine Learning
• Animal Communication

Background:

• Human expert analysis in bioacoustics can limit result reproducibility.
• Automated classification of avian vocalizations is needed to overcome these limitations.

Purpose of the Study:

• To present a machine learning method for statistically classifying avian vocalizations.
• To assess the reproducibility and reduce human bias in bioacoustic studies.

Main Methods:

• Applied automated approaches to isolate bird songs from field recordings.
• Assessed song similarities and classified songs into distinct variants using machine learning.
• Analyzed multiple replicates of automatic classification to investigate clustering uncertainty.

Main Results:

• Automatic classification of bird songs showed higher similarity to expert classifications than by chance.
• Demonstrated the presence of discrete song variants in the New Zealand hihi (Notiomystis cincta) population.
• Revealed geographic patterns of song variation by integrating classification replicates.

Conclusions:

• The developed automated approach reduces potential human bias in bioacoustic analyses.
• Facilitates greater reproducibility of results in the statistical classification of avian vocalizations.
• Highlights the utility of machine learning for understanding bird song variation and geographic patterns.