Larynx virtual surgery

Abstract

This paper presents a novel computational method designed to assist phonologists to anticipate the effects produced in the voice, as a result of physical and mechanical alterations of the larynx model, due to a surgery to correct dysphonia. The technique is based on images obtained from endoscopic exams of the larynx. The main objective of the study was to use the active contours method to develop a larynx model based on vocal fold and glottis movements. Starting with the recorded voice signal of subject, we filter the characteristics of the glottal pulse obtained from the image exam, assuming that the dysphonia is caused by irregularities of the larynx. The modifications made in the glottis model are used to estimate a new glottal filter, which is then added to the filtered signal. Using this method, one can obtain a new voice that maintains the individual's personal characteristics after virtual surgery modifications. © 2005 IEEE.