Measuring vocal difference in bird population pairs.

Area of Science:

• Bioacoustics
• Zoology
• Computational Linguistics

Background:

• Bird populations can exhibit acoustic and morphological divergence over time.
• Quantitative measures of inter-population differences are valuable for zoological studies.
• Existing speech analysis techniques can be adapted for avian vocalization analysis.

Purpose of the Study:

• To refine a speech-based dialect difference system for measuring avian vocalization differences.
• To develop a quantitative metric for assessing divergence between bird populations.
• To evaluate the effectiveness of the proposed metric across varying levels of vocal similarity.

Main Methods:

• Extracting pitch contours from bird vocalizations.
• Transposing pitch contours into pitch codes.
• Utilizing codebook schemes, including vector quantization, to represent contour structure.
• Applying the Bird Vocalisation Difference measure to diverse bird population pairs.
• Investigating the impact of dataset size on the measure's performance.
• Using Gaussian mixture models with Mel-frequency cepstral coefficients for baseline classification.

Main Results:

• The Bird Vocalisation Difference measure shows promising initial results.
• The metric's behavior aligns with established similarity levels for tested bird populations.
• Analysis indicates the influence of data size on the quantitative measure.
• Baseline classification results provide insights into class confusability.

Conclusions:

• The developed Bird Vocalisation Difference metric effectively quantifies vocal divergence in bird populations.
• The method shows potential for aiding zoologists in distinguishing between species and subspecies based on vocalizations.
• Further research can explore the application of this metric across a wider range of avian species and acoustic scenarios.