Wavelet leader multifractal analysis of period and amplitude sequences from sustained vowels

Abstract

Irregularities in the amplitude and period are characteristic of both normal and pathological sustained vowels; they are a product of perturbations inherent in the phonation process. Their analysis provides useful diagnostic information for several vocal pathologies, and their accurate modelling has been shown to improve the quality of synthesized voice. In this work, we propose the application of multifractal analysis for the characterization of amplitude and period fluctuations in sustained vowels. Using a combination of high order statistics, this signal processing tool generalizes previous approaches and provides a rich description of the fluctuation in the regularity of the data. Our results suggest that both amplitude and period fluctuations show a multifractal behavior, independent of the gender of the speaker. We also analyze the problem of classification between healthy and nonhealthy speakers as an example to show the usefulness of multifractal attributes. We conclude that amplitude and period sequences of sustained vowels should be analyzed and modelled by the multifractal paradigm.