Features Extraction on Cleft Lip Speech Signal using Discrete Wavelet Transformation

Abstract

Cleft is one of the most common birth defects worldwide, including in Indonesia. In Indonesia, there are 1,596 cleft patients, with 50.53% having a cleft lip and palate (CL/P), 24.42% having a cleft lip (CL), and 25.05% having a cleft palate (CP). Individuals with clefts encounter difficulties with resonance and articulation during communication due to dysfunctions in the oral and nasal cavi-ties. This study investigates various types of mother wavelets as feature extractors for cleft speech signals. Five different mother wavelets, namely Symlet order 2, Reverse Biorthogonal order 1.1, Discrete Meyer, Coiflet order 1, and Biorthogonal order 1.1 are analyzed. This work aims to find the best type of mother wavelet. The extracted features are statistical features, such as mean, me-dian, standard deviation, kurtosis, and skewness. The dataset used in this study consists of 200 sound signals from 10 individuals with cleft conditions and 10 normal volunteers. To assess the performance of the extractor, classification is performed using K-Nearest Neighbor (KNN) and K-Fold cross-validation. The experimental results indicate that the Reverse Biorthogonal order 1.1 mother wavelet achieves the highest accuracy compared to other types of mother wavelet, where the accuracy is 93%, with sensitivity and specificity of 94% and 92%, respectively.