Detection of the closed glottis interval

A highly reliable non-iterative algorithm for the detection of the closed glottis interval of voiced sounds is described. The method is based on an indicator of linear dependence of certain intervals of the speech signal that are considerably shorter than the pitch period. This indicator is evaluated using the normalized prediction error calculated by linear predictive coding applied to the integrated speech signal. The method allows for the decomposition of the speech signal into an excitation function and a signal for which the glottal impedance is constant and infinite. The algorithm has been extensively applied to the detection of a pitch period in the speech signal and to the extraction of the vocal tract area function. The potential applications of this approach include not only speech analysis and synthesis, but equally speech recognition and speaker identification.