Identification of Voice Disorderliness Using Machine Learning Algorithm

The speech serves as an example of a basic human behavior. Any deviation from a person's regular speaking patterns is referred to as a speech disorder. It has an impact on social integration and communication. Due to their voice dysfunction, many patients will also experience emotional and psychological problems. Affected individuals may experience a decline in their quality of life as a result of these common issues, which makes them try to avoid daily tasks. As a result, early detection of speech impairments is essential. The paralinguistic information present in human speech is essential to many speech recognition applications. One of the most crucial tasks is to detect voice disorderliness. In order to identify the disorder several features such as the ZCR, Spectral Centroid, MFCC and Mel spectrogram are taken from the speech signal. Using acoustic data taken from both male and female speech signals, this work suggests a machine learning-based method for classifying and detecting voice abnormalities. Earlier a variety of traditional (invasive) approaches were employed to diagnose speech disorders, however, much research has resulted in the creation of computer-based (non-invasive) methods. Performance models' efficacy is evaluated using evaluation criteria as accuracy, precision, recall, and F1-score. According to the findings, machine learning algorithms can greatly enhance the detection of voice problems, providing a non-invasive, economical, and effective tool for clinical decision-making and early diagnosis.