Pathological voices detection using Support Vector Machine

In our study we present a method to identity pathological voices using Support Vector Machines (SVM). Speech signals were sampled from the sustained vowel /a/ pronounced by 160 subjects (80 female and 80 male), including 80 speakers (40 women and 40 men) suffered from various dysphonias (such as acute laryngitis, adductor spasmodic, vocal fatigue, vocal tremor, vocal fold edema, laryngeal paralysis...), and 80 healthy subjects as the same previous partition. In fact, to classify pathological voices we extracted pitch from speech signal using the Fast Fourier Transform (FFT) and we estimated the three first formants by Linear Prediction Coefficients (LPC).