A robust automatic birdsong phrase classification: A template-based approach.

Area of Science:

• Bioacoustics
• Machine Learning
• Computational Auditory Scene Analysis

Background:

• Automatic bird sound phrase detection systems aid research by reducing manual annotation needs.
• Challenges include limited training data (rare phrases, few recordings) and significant background noise (wind, other animals).

Purpose of the Study:

• To present a novel template-based birdsong phrase classification algorithm.
• To develop a method robust to limited training data and noisy environments.

Main Methods:

• Utilized dynamic time-warping (DTW) and prominent time-frequency regions of training spectrograms to create templates.
• Compared the proposed algorithm against traditional DTW and hidden Markov models (HMMs) under various conditions.

Main Results:

• The proposed algorithm significantly outperforms DTW and HMMs, especially in high background noise conditions.
• Demonstrated robustness to both limited training data and environmental noise interference.

Conclusions:

• The developed template-based approach offers a robust solution for bird sound classification.
• This method is particularly effective in real-world scenarios with data scarcity and acoustic interference.