Determination of largest Lyapunov exponents of vocal signal: Application to unilateral laryngeal paralysis

It has been shown that voice signal abnormalities, particularly in unilater al laryngeal paralysis (ULP), are not always randomly distributed and that "statistical" indexes of regularity, such as jitter and shimmer, may be unr eliable in these cases. The techniques of nonlinear dynamics, particularly phase portraits, have been used to demonstrate that some of the abnormaliti es observed were the consequence of nonlinearity of glottic function. From a theoretical point of view, determination of Lyapunov exponents allows qua ntification of the complexity of the phase portraits. The authors studied v ocal signals recorded in 12 normal subjects and 26 patients with ULP and ca lculated the largest Lyapunov exponent (LLE). In normal subjects, LLE mean value was 0.380 (SD = 0.182). In patients with ULP, LLE mean value was 0.57 0 (SD = 0.337). The difference is significant at P = 0.031. Determination o f LLE was compared to more "classical" indexes such as jitter and oral airf low during phonation. A principal component analysis showed that informatio n contained in LLE was not redundant but complementary to the other paramet ers.