Acoustic Pattern Classification in Female Voice Using K-Nearest Neighbor with MFCC Feature Extraction

Abstract

This study investigates the classification of acoustic patterns in female voice signals using the K-Nearest Neighbors (KNN) algorithm and Mel-Frequency Cepstral Coefficients (MFCCs). Acoustic features derived from speech signals contain important spectral information that can be utilized to distinguish variations in voice characteristics. However, variability in speech signals and overlapping feature distributions present challenges for accurate classification. To address this issue, this study employs a structured approach comprising data preparation, MFCC feature extraction, and KNN classification. Each speech sample is represented as a 58-dimensional MFCC feature vector, and the dataset is split into testing and training subsets using a 20:80 ratio. The KNN model is trained using Euclidean distance and evaluated on precision, accuracy, recall, and F1-score. The results show that the proposed approach reaches an accuracy of 87.75%, indicating that MFCC features effectively capture acoustic characteristics in female voice signals. The confusion matrix analysis reveals that categories with distinct acoustic patterns, such as surprise and calm, achieve higher classification performance, whereas overlapping categories, such as happy and disgust, lead to increased misclassification. These findings demonstrate that KNN can serve as a reliable baseline method for acoustic pattern classification. However, further improvements can be achieved through enhanced feature representation and more advanced classification models.