An Efficient SMOTE-Based Deep Learning Model for Voice Pathology Detection

The Saarbruecken Voice Database (SVD) is a public database used by voice pathology detection systems. However, the distributions of pathological and normal samples show clear class imbalance. This study aims to develop system for classification voices that uses efficient deep learning models based on various oversampling methods, such as adaptive synthetic sampling (ADASYN), minority technique (SMOTE), Borderline-SMOTE directly applied feature parameters. suggested combinations oversampled linear predictive coefficients (LPCs), mel-frequency cepstral (MFCCs), methods can efficiently classify voices. balanced datasets from ADASYN, SMOTE, are validate evaluate models. experiments conducted using model evaluation metrics recall, specificity, G, F1 value. experimental results suggest proposed (VPD) integrating LPCs SMOTE convolutional neural network (CNN) effectively yield highest accuracy at 98.89% when classifying Finally, performances algorithms discussed. Furthermore, performance superior conventional imbalanced data algorithms, it be diagnose signals in real-world applications.