Investigating Fast Learning Network for Voice Pathology Detection

Recently, the implementation of machine learning algorithms in the analysis of voice disorders has become crucial for providing non-invasive identification of voice disorders using only audio signals. However, some voice pathology detection systems still face challenges such as working with limited acoustic databases, achieving low accuracy, and relying on constant parameters like a single number of hidden nodes. Therefore, this paper presents a method for the identification of voice pathology based on a machine learning algorithm. The voice signals are collected from a voice pathology database called the Malaysian Voice Pathology Database (MVPD). This database is created recently. The features of voices are extracted by using the Mel-Frequency Cepstral Coefficient (MFCC). Furthermore, the proposed method uses the Fast-Learning Network (FLN) algorithm for the classification part. In the proposed method, the FLN algorithm uses a different number of hidden nodes, where it starts with 20 nodes and finishes with 200 nodes. The performance of the proposed method is assessed in terms of many performance metrics. The results show that the proposed FLN algorithm achieves the highest results at the hidden nodes of 110. The FLN algorithm obtains promising results in detecting voice pathology.