Automatic classification of mice vocalizations using Machine Learning techniques and Convolutional Neural Networks.

Area of Science:

• Animal Behavior
• Bioacoustics
• Machine Learning

Background:

• Ultrasonic vocalizations (USVs) are crucial for understanding animal communication and behavioral phenotyping in research models.
• Manual classification of USVs is accurate but extremely time-consuming, hindering large-scale analysis.
• Developing automated methods is essential for efficient and standardized USV analysis.

Purpose of the Study:

• To propose and evaluate supervised learning methods for the automatic classification of USVs.
• To compare the performance of a Convolutional Neural Network (CNN) with traditional machine learning algorithms for USV categorization.
• To establish a sustainable and standardized procedure for analyzing ultrasonic communication.

Main Methods:

• USV audio tracks were segmented and manually labeled into 10 distinct classes using Avisoft software.
• A CNN was designed to classify USVs using spectrogram images as input.
• Support Vector Machine, Random Forest, and Multilayer Perceptrons were also tested using extracted numerical features.

Main Results:

• The CNN model, utilizing the full time/frequency information from spectrograms, achieved significantly higher classification performance.
• Methods relying solely on extracted numerical features showed lower performance compared to spectrogram-based approaches.
• The developed automated classification system offers a more efficient and potentially standardized alternative to manual analysis.

Conclusions:

• Supervised learning, particularly CNNs analyzing spectrograms, provides a powerful tool for automated USV classification.
• This approach enhances the efficiency and standardization of analyzing ultrasonic communication in animal models.
• The findings serve as a valuable benchmark for future research in automated bioacoustic analysis.