Comparing methodologies for classification of zebra finch distance calls.

Area of Science:

• Bioacoustics
• Animal Communication
• Machine Learning in Biology

Background:

• Bioacoustic analysis is crucial for biodiversity monitoring and understanding cognitive processes.
• Automated methods have advanced vocalization classification by species, sex, geography, and individual.
• Identifying key acoustic features is essential for real-world bioacoustic discrimination.

Purpose of the Study:

• To perform sex-based classification of zebra finch (Taeniopygia guttata) distance calls.
• To compare the effectiveness of discriminant function analyses (DFAs), support vector machines (SVMs), and artificial neural networks (ANNs) for this classification task.
• To identify the most predictive acoustic features for sex discrimination in zebra finch calls.

Main Methods:

• Utilized DFAs, SVMs, and ANNs for classification.
• Analyzed acoustic features extracted from spectrograms of zebra finch distance calls.
• Measured features such as frequency modulation, total duration, and end frequency.

Main Results:

• All three classification methods achieved high accuracy (92-96%) in sex-based call classification.
• Frequency modulation of ascending frequency, total duration, and end frequency were identified as the most predictive features.
• Results confirm the significance of call duration and frequency modulation in distinguishing male and female calls.

Conclusions:

• DFAs, SVMs, and ANNs are effective tools for bioacoustic sex classification.
• Specific acoustic features like frequency modulation and duration are critical for sex discrimination in zebra finch calls.
• The study offers a robust methodological approach for bioacoustic classification challenges.