Acoustic and perceptual analysis of vocal tremor

The objective is to present a method to analyze vocal tremor in sustained speech sounds and compare the results with a perceptual scoring of the degree of tremor. The vocal cycle lengths are tracked by salience analysis and dynamic programming. The cycle length time series is then split into three components by empirical mode decomposition: jitter, tremor and trend. Tremor cues are obtained via non-uniform quantization of the spectrum of the tremor component of the cycle length time series. The results report tremor size, tremor frequency and bandwidth of a corpus of vowels sustained by Parkinson speakers as well as their correlations with the perceived degree of tremor assessed via pairwise comparison.