FORMULATION OF A STOCHASTIC GLOTTAL SOURCE MODEL INSPIRED ON DETERMINISTIC LILJENCRANTS-FANT MODEL

GABRIEL ALEJANDRO ALZAMENDI; GASTÓN SCHLOTTHAUER; MARÍA EUGENIA TORRES

Florencia

Workshop; Models and Analysis of Vocal Emissions for Biomedical Applications (MAVEBA) International Workshop 2015; 2015

Universitá degli studi di Firenze

Accurate estimation of the glottal source is a difficult task in voice signal processing. In the past, several deterministic formulations were proposed to simulate glottal information. However, they are not able to accurately represent any perturbation or aperiodicity occurring in real voices. In this work, a glottal source model inspired on deterministic Liljencrants-Fant model and ruled by a stochastic difference equation is proposed. Following the source-filter theory, a pitch synchronous state-space voice model is formulated combining the proposed glottal model and a time-varying autoregressive vocal tract filter. State-space methods are applied for the estimation of both the glottal source and the vocal tract filter. The method here proposed proved to be useful for voice signal decomposition. Simulations with artificial voice signals demonstrated that a set of parameters characterizing the glottal source can be accurately computed. For real voices, preliminary results suggest that glottal source behavior can be suitable represented using this alternative model.