Analysis of vocal tremor and jitter by empirical mode decomposition of glottal cycle length time series

The presentation concerns a method for tracking cycle lengths in voiced speech and breaking up vocal cycle length fluctuations in cycle length jitter and cycle length tremor. The tracking of the cycle lengths is based on a dynamic programming algorithm, which does not request that the signal is locally periodic and that the average period length is known a priori. The cycle length time series are decomposed into a sum of intrinsic mode functions by means of empirical mode decomposition. These mode functions are then assigned to three phenomena, which are cycle length jitter, cycle length tremor and trend owing to intonation and physiological tremor. We report tests of the proposed analysis by means of synthetic disordered speech sounds and illustrate slow and fast cycle length perturbations in modal and essential tremor speakers.