Pseudo-periodic surrogate data in speech signals to determine intrinsic dynamics

Abstract

This paper presents the advantages of a pseudo-periodic surrogate data method to determine whether there exists an additional dynamics (nonlinear correlations) in a pseudo-periodic time series, since classic surrogate data methods used to detect nonlinearity are limited to stochastic-like data. Likewise, Lempel-Ziv complexity and Sample Entropy are introduced as discriminating statistics for null hypothesis testing. The first is based on the counting of different sub-sequences in a time series while the latter is based on a measure of signal irregularity. According to this, an effective methodology is proposed for speech signal processing using the KayPENTAX database. Experimental results showed that both statistics are able to reject the null hypothesis for the signal under analysis. Therefore, it is possible to conclude that there is an additional dynamics in the speech signals other than the pseudo-periodic behavior. Particularly, it was found that the Lempel-Ziv complexity is able to differentiate between signals with slightly different dynamics.