Speech Innovation Characterization by Higher Order Moments

Boundaries between different speech events, as closure of the glottis or abrupt changes in the vocal tract area, are detected by estimation of short-term Higher Order Moments (HOMs), as speech segments comprising two close events diverges from a gaussian model of the underlying production process: in fact, gaussian processes are completely defined by first and second order statistics, and boundaries can be detected adopting an hypotesis verification strategy based on the estimated HOMs values.