Non-intrusive technique for pathological voice classification using jitter and shimmer

Abstract

Speech signal contains two characteristics, system and source. When there is disturbance in vocal cord function, there is notable change in source characteristic. Despite the technological advances in the medical field, the voice pathologists use endoscopic methods to view the vocal cord flap movements for patients with infections and disturbances in vocal cords which are painful. This work is an alternative for classifying pathological voice from normal voice by evaluating the jitter and shimmer variations in the speech signal of an affected person. When there is any distortion in voice, it is reflected in the source characteristics. Pitch being the fundamental source characteristic, analyzing pitch helps us classify pathological voice from normal voice. Jitter and shimmer are derived characteristics of pitch. The glottal closure instants are better representatives of source compared to pitch. In this work, we have explored using the glottal closure instants to calculate the jitter, shimmer and other speech parameters instead of the pitch period. Analyzing these jitter and shimmer parameters for various pathological voices and normal voices help us to classify them. Experiments were carried out using a database containing normal and pathological voices. An accuracy of 85% was achieved for normal-pathological voice classification.