Automatic segmentation and classification of mice ultrasonic vocalizations.

Area of Science:

• Bioacoustics
• Animal Behavior
• Machine Learning

Background:

• Automatic classification of mouse ultrasonic vocalizations (USVs) is crucial for behavioral studies.
• Current methods often involve USV segmentation followed by classification.
• There is a need for improved accuracy and a higher number of discriminable USV classes.

Purpose of the Study:

• To develop an advanced system for classifying mouse USVs.
• To introduce a novel segmentation algorithm and a new set of features for USV analysis.
• To enhance the discrimination of USV classes compared to existing studies.

Main Methods:

• Developed a USV segmentation algorithm based on spectral entropy analysis.
• Extracted traditional features (time, frequency, time-frequency) and novel contour-based features from time-frequency ridge representations.
• Utilized shallow machine learning models for classification.

Main Results:

• The segmentation algorithm achieved 94% recall and 74% precision, outperforming other methods in recall.
• The classification models accurately differentiated ten different USV syllable types.
• Achieved 81.1% accuracy and 80.5% weighted F1-score in classification.

Conclusions:

• The developed system demonstrates significant improvements in both USV segmentation and classification.
• The novel approach, incorporating spectral entropy and contour-based features, enhances the analysis of mouse vocal repertoires.
• This system provides a robust tool for studying animal communication through USVs.