Identification of voice pathology from temporal and cepstral features for vowel ‘a’ low intonation

In the sphere of medicine, voice pathology analysis is a very important field. Voice makes sound as it is made up of air and vibrations of vocal cords. The wind blows across the vocal cords, forcing them to come together and vibrate when a person speaks. This vibration generates the sound. Anything that causes a change in one's voice is referred to as a voice disorder. Around 25% of the world's population suffers from voice issues as a result of their jobs requiring them to talk much louder than normal. Voice assessment of vocal pathologies is an important technique for recognising voice disorders. As the voice signal has lot of unpredictability, feature extraction will lessen the variability present. In this work three different pathologies such as Laryngitis, Leukoplakia and dysphonia are considered. Pathological voice samples pronounced with sound of vowel ‘a’ uttered in /a/_l (low pitch) is chosen from Saarbrucken voice database. Different temporal features such as signal entropy, root-mean-square, energy, zero crossing rate and Mel-Frequency Cepstral Coefficients (MFCC) are extracted from the voice signals. Results shows that among the extracted features, MFCC is able to discriminate between different voice pathologies considered.