Voice pathology detection based on the modified voice contour and SVM

In this study, a novel method based on the voice intensity of a speech signal is used for automatic pathology detection with continuous speech. The proposed method determines the peaks from the speech signal to form a voice contour. The area under the voice contour allows us to discriminate between normal and disordered subjects. In the case of disordered subjects, the calculated area under the voice contour is lower than that for a normal subject due to the malfunctioning of vocal folds, which makes the voice weaker and breathier. Some long-term features such as shimmer and jitter are based on the accurate estimation of fundamental frequency, which is itself a difficult task, especially for disordered speech signals. The proposed feature does not need to estimate the pitch period or fundamental frequency during the calculation of the voice contour and they provide a single value for the whole utterance similar to other long-term features. The voice disorder database used in this study includes 71 normal subjects and the same number of disordered subjects. Each disordered subject has one of the following voice disorders: vocal folds cysts, laryngopharyngeal reflux disease, vocal folds polyps, unilateral vocal folds paralysis and sulcus vocalis. The accuracy of the proposed method is 100%.