Usage of Time-based Glottal Parameters for the Detection of Dysphonia

The categorization of vocal issues has numerous implications in healthcare, disease treatment, and the development of innovative medical devices to assist physicians in identifying voice-related pathologies. This work utilizes the glottal signal parameters to differentiate between healthy and pathological voice, i.e., dysphonic voice. The characteristics of the glottal signal are derived using a recognized inverse filtering technique by TKK Aparat software, and they are used as inputs to a Support Vector Machine (SVM) based classifier to obtain the categorization of the voice signals in two different groups: dysphonic and normal voices. The database contains 115 audio recordings (vowel /a/) having samples corresponding to the two groups mentioned. Compared to previous research, this one achieved a classification rate of 91.3% by using a larger database for classification. The model also demonstrated strong reliability and precision, with only 10 misclassifications out of 115 samples, and achieved significant discrimination between healthy and dysphonic voices. These results highlight the effectiveness of glottal features in automated voice disorder detection.