Analysis of voice changes after thyroplasty using linear predictive coding

Abstract

Objective measurement of vocal quality is difficult in patients with severe voice disorders. Improved success has been reported using a modeling technique known as linear predictive coding. This technique uses an inverse filter to estimate a glottic excitation signal. The pitch amplitude is defined as the height of the first peak of the autocorrelation of the glottic excitation signal. In this study linear predictive coding was used to analyze voice disorders in patients with vocal fold immobility. Voice recordings were made in 16 patients undergoing vocal fold medialization and 10 patients who had no surgical procedure between measurements. The voice quality was rated by three speech pathologists. Five acoustic parameters were calculated from the samples. The best agreement with the listeners' perceptual analysis was achieved using the pitch amplitude. Both pitch amplitude and the perceptual ratings of voice quality improved in patients undergoing vocal fold medialization. Therefore the linear model of speech production and inverse filtering are useful in measuring vocal quality in patients with vocal fold immobility.