We present results of a time series analysis of glottal cycle lengths of healthy and dysphonic speakers. Even during the emission of sustained vowels by healthy speakers, glottis cycle lengths fluctuate slightly. Conventionally, the amount of jitter, as these fluctuations are commonly termed, is estimated by means of measures of dispersion of cycle lengths with respect to the average fundamental period. The problem is that measures of dispersion only are valid descriptors of jitter when perturbations of neighbouring cycles are statistically independent. This hypothesis appears not to be borne out. Therefore, we examined jitter by means of time series analysis which was able to take into account perturbation interdependencies statistically. Discriminant analysis showed that an interdependencies-related feature was indeed the most important factor among those able to distinguish between a set of healthy and dysphonic speakers. The second-most important factor was a measure of dispersion of cycle lengths from which all cycle-to-cycle interdependences had been removed by statistical processing.
Bibliographic reference. Schoentgen, Jean / Guchteneere, Raoul de (1995): "Time series analysis of glottal cycle lengths of healthy and dysphonic speakers", In EUROSPEECH-1995, 1733-1736.