Nonlinear Dynamical Analysis of Laryngeal, Esophageal, and Tracheoesophageal Speech of Cantonese

Abstract

Esophageal (SE) and tracheoesophageal (TE) speech are the most commonly used alaryngeal voicing types after total laryngectomy-a surgical procedure of removing a pathological larynx. Both SE and TE voices show more aperiodicity than normal laryngeal (NL) voices, and the vocal characteristics of alaryngeal voices are notoriously difficult to extract. The present study investigated the difference in vocal characteristics among NL, SE, and TE voices using perception measures and nonlinear dynamical analysis. Correlation dimension (D(2)) and sample entropy (SampEn) were obtained from 90 voice samples produced by 10 TE, 10 SE, and 10 NL male Cantonese speakers. Correlation between nonlinear dynamical parameters and perceptual ratings of different voices was also examined. Results show that both D(2) and SampEn values were significantly higher for TE and SE voices than NL voice. The overall perceptual judgment of SE and TE voice quality was negatively correlated with D(2) and SampEn. This finding supports the validity of using nonlinear dynamical parameters in assessing voice quality. Results of the present study also indicate that nonlinear dynamical analysis could be a supplemental tool to traditional acoustic analysis, especially for analyzing the voice quality of alaryngeal speech.