Exploiting deep neural network and long short-term memory method-ologies in bioacoustic classification of LPC-based...

Area of Science:

• Bioacoustics
• Machine Learning
• Amphibian Ecology

Background:

• Accurate identification of amphibian species is crucial for biodiversity monitoring.
• Traditional methods for amphibian identification can be time-consuming and require expert knowledge.
• Bioacoustic analysis offers a non-invasive approach to species recognition.

Purpose of the Study:

• To develop and evaluate deep learning models for classifying amphibian acoustic characteristics.
• To compare the effectiveness of different feature extraction and dimensionality reduction techniques.
• To identify the optimal combination of algorithms for amphibian bioacoustic recognition.

Main Methods:

• Data collection from 32 frog and 3 toad species in Taiwan.
• Feature extraction using Linear Predictive Coding (LPC) and Mel-frequency Cepstral Coefficients (MFCC).
• Dimensionality reduction with Principal Component Analysis (PCA) and classification using Deep Neural Networks (DNN) and Long Short-Term Memory (LSTM) on the Pytorch platform.

Main Results:

• The study successfully classified amphibian bioacoustic features using DNN and LSTM models.
• Performance metrics demonstrated the effectiveness of the developed classification system.
• Comparative analysis identified optimal feature sets and classification algorithms for amphibian species.

Conclusions:

• Deep learning models, particularly DNN and LSTM, are effective for classifying amphibian acoustic characteristics.
• The combination of LPC/MFCC feature extraction, PCA dimensionality reduction, and deep learning offers a robust approach for bioacoustic monitoring.
• This research provides a valuable tool for automated amphibian species identification and ecological studies.