VOCALEMONET: DEEP LEARNING-BASED EMOTION CLASSIFICATION IN ANIMAL SOUNDS USING MFCC FEATURES

The classification of emotional states in animal vocalizations is an emerging field with profound implications for animal welfare, veterinary diagnostics, and conservation efforts. With over 500,000 vocal samples from more than 200 species archived in repositories like the Animal Vocalization Database (AVD), the need for accurate, scalable analysis methods is critical. Traditional techniques such as spectrographic analysis and acoustic feature scoring are limited by subjectivity inter-observer agreement averages just 60% and are hindered by time-consuming processes and sensitivity to noise, which can reduce accuracy by up to 30% in field environments. To overcome these challenges, this study introduces a Deep Learning Convolutional Neural Network (DLCNN) designed to classify four distinct emotional states anger, disgust, fear, and purr based on animal vocalizations. The system employs Mel-Frequency Cepstral Coefficient (MFCC) extraction, capturing spectral features using 13 to 40 coefficients per frame, followed by noise reduction and normalization to improve robustness. Conventional models such as Support Vector Machine (SVM, 82% accuracy), K-Nearest Neighbors (KNN, 78% accuracy), Decision Tree Classifier (DTC), AdaBoost, and Linear Discriminant Analysis (LDA) have been assessed but perform inadequately with high-dimensional MFCC data and complex emotion patterns, often leading to overfitting or poor performance in noisy settings. The proposed DLCNN incorporates multiple convolutional layers to learn spatial hierarchies and dense layers for final classification, trained using categorical cross-entropy loss and the Adam optimizer. Data augmentation techniques such as pitch shifting and time stretching are applied to improve generalization. The DLCNN achieves 100% accuracy, precision, recall, and F1-score across all classes, significantly outperforming existing methods and offering a scalable, automated solution for real-time animal vocal emotion classification