A support vector machine-based voice disorders detection using human voice signal

Voice disorders are common diseases; most of the people have had experienced in their life. Voice disorder sufferers are usually not seeking medical consultation attributable to time-consuming and costly medical expenditure. Recently, researchers have proposed various machine learning algorithms for rapid detection of voice disorders based on the analysis of human voice. In this chapter, we have taken the pronunciation of vowel /a/ as the input of support vector machine algorithm. The research problem is formulated as binary classification which output will be either healthy or pathological status. Our work achieves an accuracy of 69.3% (sensitivity of 83.3% and specificity of 33.3%) which improves by 6.4%-19.3% compared with existing works. The implication of research work suggests tackling the imbalanced classification by adding penalty or generating new training data to class of smaller size. Everybody could contribute the voice signal of vowel /a/ and serving as big data pool.