Assessment of vocal dysperiodicities in disordered speech based on empirical mode decomposition

in this paper, empirical mode decomposition (eMd) is proposed as an alternative method in the framework of acoustic analysis of disordered speech for the purpose of clinical evaluation of voice. the empirical mode decomposition algorithm decomposes adaptively a given signal into oscillation modes extracted from the signal itself. the proposed approach for objective assessment of vocal dysperiodicity consists of two steps. in the first step, the dysperiodicity is estimated by using the generalized variogram. in the second step, the estimated dysperiodicity is decomposed into several narrow-band oscillating components via the eMd algorithm followed by a computation of the segmental signal-to-iMf ratio (sirseG) which is used as an acoustic marker for vocal dysperiodicity assessment. the proposed method is evaluated on a corpus comprising 251 normophonic and dysphonic speakers. results show that the acoustic marker involving some selected iMfs outperforms that obtained from a fullband analysis in terms of correlation with perceptual scores.