Vocal tremor is defined as slow modulation of fundamental frequency or its amplitude [1, 2]. Even though vocal tremor may be attributed to neurological diseases, it may also be a natural stochastic modulation of voice. Many studies try to measure these modulations assuming that they are stationary. Hence, their analysis is limited to small intervals loosing important information about vocal tremor. We propose a novel method for the estimation of the modulations which is able to adapt to nonstationary environments. The method is mainly based on an AM-FM signal decomposition algorithm which is able to estimate the instantaneous components of speech signals. Results confirm that the method successfully extract the modulations of large speech segments and robustly estimate the time-varying modulation frequency and the timevarying modulation level of vocal tremor.

Index Terms. Voice quality, Vocal tremor, AM-FM decomposition