Voice disorders assessed by (cross-) Sample Entropy of electroglottogram and microphone signals

Abstract

The quantitative analysis of vocal disorders by nonlinear signal processing methods has been extensively used in the last two decades. In this work, two algorithms for nonlinear time-series analysis, Sample Entropy and cross-Sample Entropy, are used on electroglottogram (EGG) and microphone (MIC) signals recorded from 51 normal and 80 dysphonic subjects, to obtain summary measures of voice disorders through SampEn and cross-SampEn indices. Such parameters quantify, respectively, the degree of irregularity (in the sense of self-dissimilarity) within a time-series and of asynchrony (in the sense of cross-dissimilarity) between two distinct time-series. The aims of this work are: to determine if statistically significant differences in terms of signal irregularity quantified by SampEn occur between normal and pathological subjects, investigating whether or not such differences can be equally seen in EGG and MIC; to assess if cross-SampEn reveals different degrees of asynchrony between EGG and MIC signals in the two groups. Results show that SampEn in pathological subjects is higher than in normal subjects for both EGG and MIC time-series, with a statistically significant difference detectable from both signals (Pe<10−4 for EGG and Pe<10−7 for MIC). Cross-SampEn exhibits a statistically significant difference too, showing a higher degree of cross-dissimilarity between EGG and MIC time-series for pathological subjects (Pe<10−4). In conclusion, SampEn and cross-SampEn well quantify the increase of complexity of both EGG and MIC signals and the decrease of their cross-similarity in presence of vocal disorders. Thanks to the complementarity of nonlinear indicators to the traditionally considered linear ones, SampEn and cross-SampEn appear as suitable candidates to enter the pool of approaches to investigate speech pathologies and to obtain potentially new insights on their nature.