Analysis of glottal stops using pitch synchronous integrated linear prediction residual

This work analyzes excitation source to characterize glottal stops using integrated linear prediction (ILP) residual, derived by pitch-synchronous (PS) approach. The glottal stop consonant is produced due to laryngeal gesture in the form of constricted glottis. This pressed glottal configuration, leads to period to period irregularities, aperiodicity, and asymmetry. Normalized crosscorrelation coefficient (NCC) between two successive glottal cycles is computed to capture the dissimilarity between adjacent glottal cycles. Also to characterize the variation in the abruptness of glottal pulses, ratio between strength of excitation (SoE) at two consecutive epoch locations and temporal energy distribution using waveform peak factor (WPF) are computed. To capture the asymmetric behavior of each glottal cycle, higher order statistics (HOS) measures are evaluated. These features show discrimination between glottal stop and other adjacent sounds and can therefore be used for spotting glottal stops in continuous speech.